Semantic Web 1 (2012) 1–5 IOS Press Semantically-enriched … · 2012-10-30 · Pervasive computing has matured from its origins as an academic research topic to a commercial reality

Semantic Web 1 (2012) 1–5 1IOS Press

Semantically-enriched PervasiveSensor-driven Systems 1Editor(s): Name Surname, University, CountrySolicited review(s): Name Surname, University, CountryOpen review(s): Name Surname, University, Country

Juan Ye a,∗, Stamatia Dasiopoulou b,∗∗, Graeme Stevenson a, Georgios Meditskos b, Vasiliki Efstathiou b,Ioannis Kompatsiaris b, Simon Dobson a,a School of Computer Science, University of St Andrews, UKE-mail:{juan.ye, graeme.stevenson, simon.dobson}.st-andrews.ac.ukb Information Technologies Institute, Centre of Research and Technology Hellas, GreeceE-mail:{dasiop, gmeditsk, vefstathiou, ikom}@iti.gr

Abstract. Pervasive and sensor-driven systems are by their nature open and extensible, both in terms of their inputs and thetasks they are required to perform. The data streams coming from sensors are inherently noisy, imprecise, and inaccurate, withdiffering sampling rates and complex correlations with each other: characteristics that challenge traditional approaches to storing,representing, exchanging, manipulating and programming with rich sensor data. Semantic Web technologies allow designers tocapture these properties within a uniform framework. The powerful reasoning techniques with such a representation facility haveproven to be attractive in addressing issues such as context modelling and reasoning, service discovery, privacy and trust. Inthis paper we review the application of the Semantic Web to pervasive and sensor-driven systems. We analyse the strengths andweaknesses of current and projected approaches, and derive a roadmap for using the Semantic Web as a platform on which open,standards-based pervasive, adaptive, and sensor-driven systems can be constructed.

Keywords: Semantic Web, Ontologies, Pervasive Computing, Sensor-Driven Systems, Context Modelling, Context Reasoning,Situation Recognition, Uncertainty, Provenance, Event Modelling, Semantic Service Discovery, Privacy and Trust, Programming

1. Introduction

The vision of pervasive computing was first artic-ulated by Mark Weiser in 1991: ‘the most profoundtechnologies are those that disappear. They weavethemselves into the fabric of everyday life until theyare indistinguishable from it’ [212]. Two essentialcharacteristics of this vision are that it is both sensor-

1This work has been supported by the EU FP7 projects SAPERE:Self-aware Pervasive Service Ecosystems under contract No. 256873and Dem@Care: Dementia Ambient Care – Multi-Sensing Monitor-ing for Intelligent Remote Management and Decision Support undercontract No. 288199.

*Corresponding author.**Corresponding author.

driven and adaptive. Pervasive computing assumes anenvironment saturated with new classes of intelligentand portable devices with sensing, computation andcommunication capabilities. With the help of thesedevices, a pervasive system can gather informationabout users and their surrounding environments and –without explicit instructions from users – provide cus-tomised services that adapt to the users’ current con-texts and needs. To be effective, pervasive computingmust be supported by an open and standard-based rep-resentation so as to facilitate integrating information ofheterogeneous types and modalities, as well as com-municating and exchanging information between de-vices and components.

1570-0844/12/$27.50 c© 2012 – IOS Press and the authors. All rights reserved

2 Ye et alia / Semantically-enriched pervasive sensor-driven systems

The key features of the Semantic Web, namely itsability to formally capture the intended semantics andto support automated reasoning, allow to elegantlyshare, integrate and manage knowledge coming fromheterogeneous sources, a common requirement in per-vasive, sensor-driven, and adaptive computing applica-tions. Rendering meaning explicitly, the Semantic Webfacilitates the exchange of data between systems andcomponents in an open, extensible manner, maintain-ing semantic clarity across applications. Semantic Webtechnologies have successfully addressed some perva-sive computing application concerns, such as repre-senting complex sensor data [135], recognising humanactivities [41], and modelling and querying locationdata across heterogeneous coordinate systems [179].Ontological reasoning has proven to be useful in ma-nipulating structured conceptual spaces [185,215,15].The use of ontologies, especially of upper and lowerontologies with appropriate connections [177], power-fully supports co-operation of data sources within anopen system.

However, the potential of Semantic Web technolo-gies to address other key pervasive computing applica-tion requirements is yet to be fully explored. Open is-sues include capturing the temporal semantics of data;reasoning in the presence of extensive uncertainty;querying and applying different reasoning schemes tohighly dynamic data; capturing provenance – and in-tegrating all these elements within a single, coherentframework. In this paper we explore these open issuesand seek to answer three questions for each: to whatextent do existing Semantic Web technologies addressthe requirements?, what additional techniques mightbe needed?, and how might the research communityaddress these deficiencies?.

We structure our discussion as follows. Section 2provides a background to pervasive sensing systems,including representative examples of applications andthe underlying research themes. It further describesand analyses the characteristics and relationships of‘typical’ information that needs to be modelled, in-cluding sensor data, context, events, domain knowl-edge and situations. Section 3 briefly introduces theSemantic Web and highlights the key enabling featuresfor knowledge capture and reasoning in the open envi-ronment of pervasive systems. Section 4 describes rep-resentative ontology-based models for pervasive com-puting, showing how and where Semantic Web tech-nologies are employed. Section 5 analyses the use ofSemantic Web technologies in terms of key pervasiveapplication requirements, including modelling of sen-

sor data, context (information about when, where, who,and what, events), reasoning towards higher-level con-text abstractions, uncertainty handling, semantic ser-vice discovery, and privacy and trust. Section 6 identi-fies open research issues in pervasive computing whichmay potentially be addressed through the use of Se-mantic Web technologies. Section 7 concludes the pa-per.

2. Background of Pervasive Computing

Pervasive computing aims to enable and assist peo-ple in accomplishing an increasing number of per-sonal and professional transactions using intelligentand portable appliances, devices, and artefacts that al-low them to employ intelligent networks and gain di-rect, straightforward, and secure access to both rele-vant information and services. It gives people access toinformation stored across potentially large-scale staticand ad-hoc networks, allowing them to easily take ac-tions anywhere and at any time. In the following, weintroduce representative application areas and researchthemes in pervasive computing and elicit requirementsof building a pervasive system. We also describe thetypical information needs of a pervasive system anddiscuss the requirements and challenges to modellingand reasoning on it.

2.1. Example Applications

Pervasive computing has matured from its originsas an academic research topic to a commercial reality.Its potential applications range from intelligent officesto smart homes, healthcare, transportation, and envi-ronment monitoring. Here we outline example applica-tions to provide a general pervasive computing back-ground.

2.1.1. Healthcare and Smart HomesPervasive healthcare is an emerging research dis-

cipline, focusing on the development and applicationof pervasive computing technologies for healthcareand wellbeing. This research is driven by the pressingneeds of an expanding ageing population, a predictedshortage in caregivers for the elderly, and the matura-tion of sensing and communication technologies [142].

Pervasive computing technologies introduce new di-agnostic and monitoring methods that directly con-tribute to improvements in care, therapy and medicaltreatment. These examples involve sensors and mon-

Ye et alia / Semantically-enriched pervasive sensor-driven systems 3

itoring devices that collect and disseminate informa-tion, such as heart rate, blood pressure and glucose lev-els, to healthcare providers [21]. They support a bet-ter understanding of facets of patients’ daily lives, sup-porting appropriate adaptation of therapies to the in-dividual. One scenario is a hospital where a patient isconstantly monitored, and findings linked to a diag-nostic process. Thus, it is possible to advise the hos-pital canteen to prepare special food for this particu-lar patient and to adapt the patient’s specific medica-tion according to his current health condition. Perva-sive computing technologies can also improve medi-cal treatment procedures. In emergency care, they canaccelerate access to medical records at the emergencysite or seek urgent help from multiple experts virtually.In the surgical field, they can collect and process anever-increasing range of telemetric data from instru-ments used in an operating room and augment humanability to detect patterns that could require immediateaction [24].

To maintain the health and safety of the home andresidents and maximise the comfort of the residents,researchers deploy sensors in a home setting to per-ceive the state of the environment and its residences,and reason about the state using artificial intelligencetechniques [49]. The DOMUS lab, at the Universityof Sherbrooke, achieves an implementation of smarthome based pervasive assistants which act as mobileorthosis [50]. Its research goal is to collect ecologi-cal data on symptoms and medication side-effects oncognitively impaired people, establish their behaviourmodels, allow caregivers to monitor their daily activi-ties, and provide assistance adapting to different typesof cognitive deficits such as memory, initiation, plan-ning, and attention [25].

2.1.2. TransportationPervasive computing technologies are entering our

everyday life as embedded systems in transporta-tion [69,53]. A number of applications have emerged.In tourist guides, a pervasive computing system canprovide personalised services (like locating a particu-lar type of restaurant or planning a day trip) for visi-tors based on their location and preferences [225]. Intraffic control, a system can be immediately informedof incidences of congestion or the occurrence of acci-dents and notify all approaching drivers. In route plan-ning, a system can suggest the most convenient routesfor users based on the current traffic conditions and thetransportation modes being used [223].

GPS-enabled devices like smartphones are chang-ing the ways in which people interact with the Web byusing their locations as context. Multiple users recordtheir outdoor movements as GPS trajectories for travelexperience sharing, life logging, sports activity analy-sis, and multimedia content management. This resultsin a large accumulation of GPS trajectories that areaccumulating unobtrusively and continuously in Webcommunities [225]. One use of these data is to discoverinteresting locations or location sequences as touristrecommendations. Here ‘interesting’ means that a lo-cation is not only visited frequently by tourists but alsois a meaningful place such as somewhere with cul-tural significance, a shopping mall, a restaurant, or abar [225].

2.1.3. Environment MonitoringAdvances in sensing and wireless network technolo-

gies provide realistic solutions for continuous moni-toring of natural environments. They potentially pro-vide new data for environmental science as well as vi-tal hazard warnings [83]. They allow the collection offine-grained environmental data, which is essential tofoster scientific research and to increase human un-derstanding of the environment. For example, detailedair pollution data collection helps to understand pol-lutants’ spreading mechanisms and their influence onhuman health [167]. In agriculture, sensors and actua-tors operating at a much finer level of granularity canbe used to precisely control, for example, the concen-tration of fertiliser in soil based on information gath-ered from the soil itself, the ambient temperature, andother environmental factors [76].

Pervasive sensing technologies are also particu-larly important in remote, inhospitable, or danger-ous environments where fundamental processes haverarely been studied due to their inaccessibility [116].Some examples are: monitoring a chemical indus-trial environment to minimise the hazards of chemi-cal wastage [109], monitoring gas emissions at land-fill sites [103], and studying active volcanoes with col-lected seismic and infrasonic (low-frequency acoustic)signals.

2.1.4. SummaryBeyond the above three types of applications, per-

vasive computing has also been applied to other ar-eas, including workplaces, education, and office envi-ronments [50]. In education, intelligent classrooms areprototyped to enhance the classroom experience by au-tomating processes such as controlling light settings,playing videos, and displaying slides [221]. Smart


meeting applications facilitate archiving, analysing,and summarising meetings via a variety of technolo-gies from physical capture from camera, microphone,or other interactive tools and structural analysis to se-mantic processing [72].

2.2. Research Themes

Pervasive sensing systems offer a framework fornew and exciting research across the spectrum of com-puter science. Major research themes cover the follow-ing areas [188]:

– Sensing and reasoning In order to develop reac-tive pervasive computing systems it is necessaryto capture, process, and exploit the contextual pa-rameters that inform and guide human behaviour.

– Privacy, trust, and security Privacy encompassesreasoning about trust and risk involved in the in-teractions between users and services. Trust con-trols the amount of information that can be re-vealed in an interaction. Risk analysis allows us toevaluate the expected benefit that would motivateusers to participate in these interactions. Securitydescribes the cryptographic techniques used to se-cure the communication channels and participantdata.

– Discovery One of the key requirements for per-vasive computing systems is an approach or ser-vice capable of assimilating and filtering informa-tion from its inputs (such as sensors, services, ap-plications, and users). Given some infrastructureto communicate this information, an approach tomatching that correlates relevant input events andfacts to a particular contextual service is needed.This is essential to allow the user and the applica-tion to discover the information from the environ-ment necessary to achieve a defined goal or com-plete a task.

– Interaction design As traditional computers dis-appear from pervasive computing environments,novel human-computer interactions to handle thepeculiarities of their environments – including in-visible devices, implicit interaction, and real, vir-tual, and hybrid interactions – need to be investi-gated.

– Essential infrastructure It is necessary to developtools to maintain and upgrade infrastructure in-place over its lifecycle, as well as to allow theinfrastructure to communicate failures effectivelyto its users.

2.2.1. SummaryWe have briefly introduced the background of per-

vasive and sensor-driven computing in terms of appli-cations and research themes. From this, we concludethat data and knowledge modelling is key in pervasivecomputing. We summarise the following requirementsthat are particularly relevant to the Semantic Web tech-nology community.

– Conceptual modelling: We require a commonly-agreed vocabulary to represent data so as to makethem understandable across different systems andplatforms; e.g., populating GPS or social data inWeb communities, or transferring health recordsacross hospitals and medical organisations. Sucha vocabulary should be rich enough to representproperties, structures, and relationships amongdata, thus facilitating querying, personalising andfurther processing of data, as well as their sharingand reuse. It will be used to semantically annotatedata so not only are they represented in machine-processable formats but they also are ‘meaning-fully’ related to other data; e.g., in the GPS exam-ple in Section 2.1.2, a place needs to be annotatednot only as a location but also related to culturalor social aspects.

– Reasoning: Reasoning is a key enabler for perva-sive applications in that there is a large amountof information in broad types that needs to be fil-tered and abstracted to higher-level informationthat is understandable or interesting to applica-tions or humans; e.g., inferring a patient’s condi-tion based on symptoms acquired from biochemi-cal sensors, or analysing air pollution data to pre-dict the pollution spreading direction so as to takeimmediate action to mitigate its effects.

– Uncertainty: Context information is inherentlyimperfect. Inaccuracies may easily arise, due toerroneous and/or missing readings. Furthermore,when pieces of context information come frommultiple sources and modalities, ambiguities andconflicts may also arise. Under these circum-stances, modelling and reasoning need to providethe means to cope with such imperfections andallow to detect possible errors, handle gracefullymissing values, and derive plausible conclusions,assessing the validity of available context data.

– Service discovery: Pervasive and sensor-drivensystems are usually assumed to operate in large-scale, open environments where components areloosely coupled and have volatile presence; i.e.,


they can enter or leave a system freely at anypoint. Thus, it is desirable to associate them withsemantic annotations so as to support matching,discovery, composition, and orchestration in amore intelligent manner.

– Privacy and trust: Data in pervasive systems aretypically associated with privacy concerns; espe-cially personal data such as health records. Weneed a specification language to specify rules orpolicies about how to access, appropriately (per-haps automatically) abstract, and use data.

In Section 4, we introduce Semantic Web-based per-vasive systems and indicate which of these require-ments are favourably supported. In Section 5, we anal-yse in detail the extent to which these requirements aresatisfied.

2.3. Information in Pervasive Sensing Systems

We introduce the types of information generated ina pervasive sensing system and highlight modellingand reasoning challenges. Figure 1 lists informationincluding raw sensor data, events, context, domainknowledge, and situations or human activities.

Raw Sensor Data

High-level Event Human Activity Situation

ContextLow-level Event Knowledge

Fig. 1. Different types of information in pervasive sensing systems

Raw sensor data are readings reported from sensors;e.g., a complete reading from Ubisense – a coordinate-based positioning sensor is ([4.54,1.92,4.70],tag_bob, 09/05/2008 09:49:39), includingwhen the reading is generated (e.g., 09/05/200809:49:39), who the reading is about (e.g., a tag be-longing to the user named bob), and what the readingvalue is (e.g., a 3D coordinate-based location datum[4.54,1.92,4.70]).

Context is defined by Dey et al. [57] as ‘any infor-mation that can be used to characterise the situationof an entity’. We regard context as a well-structured

concept that describes a property of an environmentor a user. For example, a location context can berepresented by means of the Coordinate, Geometri-cArea and SymbolicLocation classes. GeometricAreainstances describe areas that map to a set of Coordi-nate instances (numerical value array); SymbolicLoca-tion instances assign a human-friendly symbol to Ge-ometricArea instances.

Knowledge here indicates a corpus of knowledgespecific to available context, including spatial knowl-edge for location contexts, or user preference and so-cial network for person contexts.

An event represents a change in the state of an envi-ronment that should be identified, processed and man-aged by the system in order to deliver personalised ser-vices to users; e.g., an event of ‘a user having heart at-tack’ might trigger an application of calling an ambu-lance. A low-level event can be conceptually defined asan action or occurrence that happens at a certain timewithin an environment, which is described through ba-sic elements such as when, where, what, and who. Ahigh-level event includes other parameters includingrole and factor.

Situations are defined as an abstraction of the statesoccurring in the real world, which are usually inferredfrom the aggregation of multiple contexts. They mayrepresent a human activity (e.g., a user is ‘having agroup meeting’) or a state of affair (e.g., there is an on-going presentation happening in a room). Situations,human activities, and high-level events are used inter-changeably in the community of pervasive computing,depending on the application context. In the following,we will discuss the features of the above types of in-formation in detail.

2.3.1. Sensors and Sensor DataThe service provision of a pervasive computing sys-

tem relies on the perception of an environment, whilethe perception is accomplished by a range of sensors.Sensors are becoming smaller, lighter, lower in cost,and aided by longer battery life. These advances in-creasingly support the embedding of sensors withinan environment, integration with everyday objects, andattachment to the human body. Sensors in pervasivecomputing can capture a broad range of informationon the following aspects [51]:

– Environment: temperature, humidity, barometricpressure, light, and noise level in an ambient envi-ronment and usage of electricity, water, and gas;

– Device: state of devices (e.g., available or busy),functions of devices (e.g., printing or photocopy-


ing), the size of memory, the resolution of screen,or even operating systems embedded;

– User: location, schedule, motion data like accel-eration of different parts of bodies, and biometricdata like heart rate and blood pressure;

– Interaction: interactions with physical artefactsthrough RFID and object motion sensors [114],and interactions with devices through virtual sen-sors like monitoring the frequency a person usestheir keyboard and mouse [133,123].

The diversity of sensors leads to high complexityin their output, including volume, different modalities,inter-dependence, real-time update, and critical age-ing. In dealing with the real world, these sensors donot always produce accurate data. Noisy sensor datamay result in misunderstanding of a user’s or an envi-ronment’s state, which may lead to incorrect applica-tion behaviour. Sensors have technical limitations, areprone to breakdown, can become disconnected fromthe sensor network, or may be vulnerable to environ-mental interference. This leads to issues surroundingthe uncertainty of sensor data, which are out of date,incomplete, imprecise, and contradictory [84]. Thesefeatures of pervasive sensor data can be an impedimentto the process of making them directly usable or un-derstandable by applications.

Different sensors produce data of different modal-ities, including binary, continuous numeric, and fea-tured values. The type of data has an impact on choos-ing appropriate structure to represent it so as to makeit understandable and shareable among different plat-forms and systems. A binary value is the simplest typeof sensor data: true (1) or false (0). RFID sensors pro-duce a binary reading: an object with a RFID tag isdetected by a reader or not; binary-switch sensors be-have similarly [205]. Continuous numeric values are acommon type of data that is produced by most sensors,including positioning sensors, accelerometers, and allthe ambient sensors. Featured values are typically pro-duced from sophisticated sensors such as a camera andan eye movement tracker whose data needs to be char-acterised into a set of categorical measurements. Forexample, motion features can be extracted from videostreams recorded in cameras, including quantity of mo-tion and contraction index of the body, velocity, accel-eration and fluidity [31]. Eye movements captured inelectrooculography signals are characterised into twotypes of features: saccades that refer to the simultane-ous movement of both eyes in the same direction andfixations that are the static states of the eyes duringwhich gaze is held upon a specific location [26].

2.3.2. Context, Events and SituationsSensor data can be abstracted into a set of domain

concepts, typically called context. Contexts can beclassified into different domains in terms of the prop-erty they describe. Contexts in one domain can have adifferent structure, which is distinguishable from con-texts in other domains. For example in the domain ofLocation a coordinate context can be identified as threenumeric values with their unit of measurement as Me-ter, while in the domain of Temperature a context canbe identified as a numeric value with its unit of mea-surement as Celsius or Fahrenheit. A higher-level con-text can be considered as a characteristic function onsensor data; for example, a symbolic location contextstudyRoom is mapped to a set of coordinate contextsthat are valid inside a study room. An event translatesor abstracts sensor data into a relation with contexts.

Semantic relationships exist between contexts inone domain, including different levels of granularity,overlapping, and conflicting [218,170]. The seman-tic relationships can be explored by applying domainknowledge or through a learning process. For example,given a map of a house, we can specify the spatial rela-tionships between the location contexts as: the locationcontext studyRoom is finer grained than another lo-cation context house (ref. spatially contained in), andit conflicts with the context bedroom (ref. spatiallydisjoint with).

Events are regarded as occurrences upon which ac-tions are taken. They are considered enduring entitiesthat unfold over time and space; that is, they occur orhappen at certain time in certain space [168]. An eventmay be complex and have a variety of aspects beyondtime and space, including objects and people involved,and as well as mereological, causal, and correlative re-lationships between them [168].

A situation is defined as an external semantic in-terpretation of sensor data [217]. Interpretation meansthat situations assigns meanings to sensor data. Exter-nal means that the interpretation is from the perspec-tive of applications, rather than from sensors. Semanticmeans that the interpretation assigns meaning on sen-sor data by defining structure or relationships withinthe same type of sensor data and between differenttypes of sensor data.

A situation or human activity can be defined by col-lecting relevant contexts, uncovering meaningful cor-relations between them, and labelling them with a de-scriptive name. The descriptive name can be calleda descriptive definition of a situation, which is abouthow a human being defines a state of affairs in reality.


A logical expression of correlated context predicatesis called a logical specification of a situation. Withthese two, a situation bridges sensor data and applica-tions. Sensor data are abstracted to a certain situationby evaluating its specification, and this situation willtrigger applications that correspond to its descriptivename.

There are rich structural and temporal relationshipsbetween situations:

Temporal feature The occurrence of a situation usu-ally exhibit certain temporal features: time-of-day– a situation may only happen at a particular timeof the day; duration – it may only last a certainlength of time; frequency – it may only happen acertain number of times per week, and sequence– different situations may occur in a certain se-quence. A situation may occur before, or afteranother situation, or interleave with another sit-uation; for example, ‘taking pill’ should be per-formed after ‘having dinner’ [97].

Generalisation A situation can be regarded as moregeneral than another situation, if the occurrenceof the latter implies that of the former; for ex-ample, a ‘watching TV’ situation is consideredmore specific than an ‘entertainment’ situation,because the conditions inherent in the former sit-uation subsume or imply the conditions in the lat-ter situation [216].

Composition A situation can be decomposed into aset of smaller situations, which is a typical com-position relation between situations. For exam-ple, a ‘cooking’ situation is composed of a ‘us-ing stove’ situation and a ‘retrieving ingredients’situation. McCowan et al. propose a two-layeredframework of situations: a group situation (e.g.,‘discussion’ or ‘presentation’) is defined as acomposition of situations of individual user (e.g.,‘writing’ or ‘speaking’) [119].

Dependence A situation depends on another situa-tion if the occurrence of the former situation isdetermined by the occurrence of the latter situ-ation. Dependence can be long- or short-range,as proposed by [42]. Sometimes long-range de-pendence can be more useful in inferring high-level situations. For example, a situation ‘going towork’ may be better in inferring a situation ‘go-ing home from work’ than other short-range de-pendent situations.

2.4. Summary

As sensing technologies mature, pervasive comput-ing is increasingly being deployed in real-world set-tings. To facilitate interoperability between sensors,reasoners, and applications (that is, to support a sharedunderstanding and a standard approach to query-ing), we need a sensor model that provides machine-readable specifications of sensors, their output types,and the domains in which they operate so that it cansupport classifying sensors according to their function-alities, finding a sensor that can perform a certain mea-surement, or even composing existing sensors to createa virtual sensor for particular functionality [48]. Thesensor model also needs to capture information aspectsabout observation data such as when, what and qual-ity measures (e.g., sampling frequency, coverage area,precision or accuracy), as well as to enable the deriva-tion of high-level knowledge from low-level sensordata.

The explicit modelling of events is beginning to gainwidespread interest, as the increasing number of sen-sors leads to an ubiquity of events that need to berecognised and communicated. Event detection, clus-tering and annotation are the major concerns in eventmodelling.

Context- and situation-awareness is a key enabler ofpervasive systems; that is, the behaviour of the servicesprovided, depends on a system’s ability to respond tousers’ current contexts or situations. We need a contextand situation model with rich semantics to express un-certainty, temporal features, generalisation, composi-tion, and dependence. Such a model also requires pow-erful reasoning capabilities to effectively represent andefficiently reason on these properties and relationships.

3. Semantic Web Technologies

The Semantic Web [14] is a resource-oriented ex-tension of the current Web that aims to afford a com-mon framework for sharing and reusing data acrossheterogeneous agents, applications and systems. Theunderlying rationale is to make the semantics of webresources explicit by attaching to them metadata thatdescribe meaning in a formal, machine-understandableway. Within this vision, ontologies play a key role, pro-viding consensual and precisely defined terms for thedescription of resources in an unambiguous manner.

In 2004, the Web Ontology Language (OWL) [204]became a W3C recommendation, paving the way


for the development of adequate tool support (on-tology editors, reasoners, etc.) and the proliferationof ontology-based applications in a plethora of do-mains. Formally founded in Description Logics (DLs)[9], OWL is endowed with expressive representationalconstructs that allow to capture complex knowledge.At the same time, OWL avails of the well-defined DLreasoning services for affording automated reasoningsupport.

The aforementioned furnish OWL with a multiplic-ity of appealing, within the context of pervasive appli-cation features. For example, in OWL one can effec-tively model and reason over taxonomic knowledge.This is a desirable feature in pervasive applicationswhere the need to model information at different lev-els of granularity and abstraction, so as to drive thederivation of successively further detailed contexts isparticularly evident. Similarly, OWL supports consis-tency checking, another useful feature when dealingwith imperfect context information coming from mul-tiple sources.

Last but not least, OWL adheres to the open-worldassumption, providing an elegant way of modelling in-complete information. Intuitively, open-world seman-tics assumes that we do not have complete informa-tion about the world, but that some information maybe missing. Such assumption is well-suited for sensor-driven systems, where information may be incompletedue to a number of reasons, such as sensor inaccu-racies or imperfect observations as described in Sec-tion 2.3.1.

3.1. Description Logics

The design of OWL, and particularly the formalisa-tion of the semantics and the choice of language con-structors, have been strongly influenced by Descrip-tion Logics (DLs) [93]. DLs are a family of knowl-edge representation formalisms characterised by logi-cally grounded semantics and well-defined reasoningservices [9].

The main building blocks are concepts represent-ing sets of objects (e.g., Person), roles representingrelationships between objects (e.g., worksIn), and in-dividuals representing specific objects (e.g., Alice).Starting from atomic concepts, such as Person, ar-bitrary complex concepts can be described througha rich set of constructors that define the conditionson concept membership. For example, the concept∃hasFriend.Person describes those objects that are re-lated through the hasFriend role with an object from

Table 1Terminological and assertional axioms

Name Syntax Semantics

Concept inclusion C v D CI ⊆ DI

Concept equality C ≡ D CI = DI

Role equality R ≡ S RI ⊆ SI

Role inclusion R v S RI ⊆ SI

Concept assertion C(a) aI ∈ CI

Role assertion R(a, b) (aI , bI) ∈ RI

the concept Person; intuitively this corresponds to allthose individuals that are friends with at least one per-son.

A DL knowledge base K typically consists of aTBox T (terminological knowledge) and an ABox A(assertional knowledge). The TBox contains axiomsthat capture the possible ways in which objects of a do-main can be associated. For example, the TBox axiomDog v Animal asserts that all objects that belong tothe concept Dog, are members of the concept Animaltoo. The ABox contains axioms that describe the real-world entities through concept and role assertions. Forexample, Dog(Jack) and isLocated(Jack,kitchen) ex-press that Jack is a dog and he is located in the kitchen.Table 1 summarises the set of terminological and as-sertional axioms.

The semantics of a DL language is formally de-fined through an interpretation I that consists of a non-empty set ∆I (the domain of interpretation) and an in-terpretation function .I , which assigns to every atomicconcept A a set AI ⊆ ∆I and to every atomic role R abinary relation RI ⊆ ∆I ×∆I . The interpretation ofcomplex concepts follows inductively. Table 2 showsthe syntax and semantics of some of the most commonDL constructors.

3.2. Reasoning Services

Besides formal semantics, DLs come with a setof powerful reasoning services, for which efficient,sound and complete, reasoning algorithms, with well-understood computational properties, are available(e.g., tableaux-based algorithms [10]). Example state-of-the-art implementations include Pellet [174], Racer[82], Fact++ [200] and Hermit [132].

Assuming a DL knowledge base K = (T,A), typicalreasoning services include:

– Subsumption: A concept C is subsumed by D inT (written C v D), iff CI ⊆ DI for all interpre-tations I.


Table 2Examples of concept and role constructors

Name Syntax Semantics

Top > ∆I

Bottom ⊥ ∅Intersection C uD CI ∩DI

Union C tD CI ∪DI

Negation ¬C ∆I \ CI

Universal quant. ∀R.C{a ∈ ∆I | ∀b.(a, b) ∈ RI

→ b ∈ CI}

Existential quant. ∃R.C{a ∈ ∆I | ∃b.(a, b) ∈ RI

∧ b ∈ CI}

Inverse R−{(b, a) ∈ ∆I ×∆I |

(a, b) ∈ RI}Transitive closure R+

⋃n≥1(RI)n

Composition R ◦ S RI ◦ SI

– Equivalence: Two concepts C and D are equiva-lent in T (written C ≡ D) iff C v D and D v C.

– Disjoint: A concept C is disjoint to a concept D inT iff in every interpretation I it holds that CI 6=∅.

– Satisfiability: A concept C is satisfiable in T iffthere is an interpretation I such that CI 6= ∅.

– Consistency: The ABox A is consistent w.r.t. T iffif there is an interpretation that is a model of bothA and T .

– Instance checking: The individual a is an instanceof C w.r.t. (K) (written K |= C(a)) iff aI ∈ CI

holds for all interpretations I of K– Realisation: An instance’s a realisation w.r.t. toK includes finding the most specific concepts Cfor which aI ∈ CI holds for all interpretations Iof K.

Hence, through subsumption one can derive the im-plicit taxonomic relations among the concepts of a ter-minology. For example, given the axiom Occupied-Room v Room u ∃contains.Person, one can infer thatRoom subsumes OccupiedRoom.

Satisfiability and consistency checking are useful todetermine whether a knowledge base is meaningful atall. Satisfiability checking enables the identification ofconcepts for which it is impossible to have membersunder any interpretation (for example, an unsatisfiableconcept, though trivial, is OccupiedRoom u ¬ Occu-piedRoom). Consistency checking enables the identi-fication whether the set of assertions comprising theknowledge base is admissible with respect to the termi-nological axioms. For example if EmptyRoom and Oc-cupiedRoom are asserted as disjoint concepts, then the

presence of both OccupiedRoom(kitchen) and Empty-Room(kitchen) leads to inconsistency.

Instance checking denotes the task of finding whethera specific individual is an instance of a given concept.Realisation of an individual, a more generic form ofinstance checking, returns all (most specific) conceptsfrom the knowledge base that a given individual is aninstance of. Its dual is the retrieval problem that givena specific concept C it returns all individuals that be-long to this concept. This reasoning service is the cen-tral one to realise the task of recognition of situationtypes.

A particularly appealing feature of DLs is that theyare decidable fragments of first-order logic. Their de-cidability is primarily due to the so called tree modelproperty [206], according to which a class C has amodel (i.e. an interpretation I in which CI is non-empty) iff C has an interpretation that defines a tree-shaped directed graph. As we later discuss, this prop-erty has a direct impact on the relational expressivity ofDLs and OWL, and has been among the principal mo-tivating factors for investigating the integration withrules.

Falling under the Classical logics paradigm [144],reasoning in DLs (and hence in OWL) adopts theopen-world assumption. Intuitively, if a fact a holdsonly in a subset of the models of the knowledge baseKB, then we can conclude neither KB |= a nor KB6|= ¬a. For example, if the only available knowl-edge regarding the residents of a house is the asser-tion livesIn(Alice,house), we cannot deduce based onit alone that no one else lives in the house. In con-trast, formalisms adhering to the closed-world assump-tion make the common-sense conjecture that all rele-vant information is explicitly known, so all unprovablefacts should be assumed not to hold. In our example,this amounts to concluding that Alice is the sole resi-dent of this house. Hence, closed-world reasoning canbe intuitively understood as reasoning where from KB6|= a one concludes KB |= ¬a [129]. Such kind rea-soning should not be confused however with closed-domain reasoning, which has also been used in a num-ber of context reasoning frameworks (Section 5.4), andwhich involves reasoning only over explicitly knownindividuals.

3.3. OWL and OWL 2

OWL comes in three dialects of increasing expres-sive power: OWL Lite, OWL DL and OWL Full [7,91].The first two languages can be considered as syntactic


variants of the SHIF(D) and SHOIN (D) DLs re-spectively. The third language was designed to providefull compatibility with RDF(S), which is the most ex-pressive of the three. It neither imposes any constraintson the use of OWL constructs, nor lifts the distinctionbetween instances (individuals), properties (roles) andclasses (concepts). This high degree of expressivenesscomes however at a price, namely the loss of decid-ability that makes the language difficult to implement.As a result, focus has placed on the two decidable di-alects, and particularly on OWL DL, which is the mostexpressive of the two.

Despite the rich primitives provided for expressingconcepts, OWL DL has often proven insufficient to ad-dress the needs of practical applications. This limita-tion amounts to the Description Logic style model the-ory used to formalise its semantics, and the tree-modelproperty role in conditioning DLs decidability. As aconsequence, OWL can model only domains whereobjects are connected in a tree-like manner. This con-straint can be restrictive for real-world applications,including the pervasive domain, which requires mod-elling general relational structures (see Section 5.4).

Responding to this limitation and to other draw-backs that have been identified concerning the use ofOWL in different application contexts throughout theyears, the W3C working group produced OWL 2 [77].OWL 2 (equivalent to the SROIQ(D) DL) is a re-vised extension of OWL, now commonly refereed toas OWL 1. It extends OWL 1 with qualified cardinal-ity restrictions; hence one can assert that a social ac-tivity is an activity that has more than one actors: So-cialActivity v Activity u ≥ 2 hasPar-ticipant.Person. Furthermore, it makes it possi-ble to define properties to be reflexive, irreflexive, tran-sitive, and asymmetric, and to define disjoint pairs ofproperties, thus providing extended support for captur-ing mereology relations.

Another prominent OWL 2 feature is the extendedrelational expressivity that is provided through theintroduction of complex property inclusion axioms(property chains). To maintain decidability, a regular-ity restriction is imposed on such axioms that dis-allow the definition of properties in a cyclic way.Hence, one can assert the inclusion axiom locatedIn◦ containedIn v locatedIn making it possible to in-fer that if a person is located for example in the En-gineering Department of the University, then she islocated in the University as well; however, it is notallowed to use both the aforementioned axiom andthe axiom containedIn ◦ locatedIn v containedIn

as this leads to a cyclic dependency. Three profiles,namely OWL 2 EL, OWL 2 QL and OWL 2 RL, tradeportions of expressive power for efficiency of reason-ing targeting different application scenarios.

3.4. Combining Ontologies and Rules

To achieve decidability DLs, and hence OWL, tradesome expressiveness for efficiency of reasoning. Thetree-model property, mentioned above, is one such ex-ample. It conditions the tree-shape structure of models,ensuring decidability, but at the same time it severelyrestricts the way variables and quantifiers can be used,dictating that a quantified variable must occur in aproperty predicate along with the free variable. As aresult, it is not possible to describe classes whose in-stances are related to an anonymous individual throughdifferent property paths [78]. To leverage OWL’s lim-ited relational expressivity and overcome modellingshortcomings that OWL alone would be insufficient toaddress, a significant body of research has been de-voted to the integration of OWL with rules.

A proposal towards this direction is the SemanticWeb Rule Language (SWRL) [92], in which rules areinterpreted under the classical first order logic seman-tics. Allowing concept and role predicates to occur inthe head and the body of a rule without any restrictions,SWRL maximises the interaction between the OWLand rule components, but at the same time rendersthe combination undecidable. To regain decidability,proposals have explored syntactic restrictions on rules[162,131] as well as their expressive intersection [78].The DL-safe rules introduced for example in [131] im-pose that rule semantics apply only over known indi-viduals. It is worth noting that in practice DL reason-ers providing support for SWRL actually implement asubset of SWRL based on this notion of DL-safety.

Parallel to these efforts, a highly challenging andactive research area in the Semantic Web addressesthe seamless integration of open and closed world se-mantics. Representative initiatives in this quest includeamong others the hybrid formalism of MKNF knowl-edge bases proposed by Motik and Rosati [130], theextension of ontologies through the use of integrityconstraints proposed by Tao et al. [190], and the so-called grounded circumscription approach [108].

Taking a different perspective, a number of ap-proaches have investigated the combination of ontolo-gies and rules based on mappings of a subset of the on-tology semantics on rule engines. For instance, Horst[193] defines the pD* semantics as a weakened variant


of OWL Full, e.g., classes can be also instances, andin [194] they are extended to apply to a larger subsetof the OWL vocabulary, using 23 entailments and 2 in-consistency rules. Inspired by the pD* entailments andDLP [78], the semantics of the OWL 2 RL profile isrealised as a partial axiomatisation of the OWL 2 se-mantics in the form of first-order implications, knownas OWL 2 RL/RDF rules. User-defined rules on topof the ontology allow to express richer semantic re-lations that lie beyond OWL’s expressive capabilities,and couple ontological and rule knowledge. Examplesof rule-based OWL reasoners that implement entail-ment rules, include among others OWLIM [18], OWL-JessKB1, Jena [117], O-DEVICE [124] and DLEJena[125].

3.5. Summary

This section introduced the basic notions underly-ing the Semantic Web and provided a brief overviewof key technologies empowering the envisaged knowl-edge sharing and reuse across heterogeneous environ-ments. Expressive ontology languages allow to ele-gantly capture complex knowledge and its semanticsin a formal way, rendering it amenable to automatedreasoning tasks with well-understood computationalproperties. Rules augment further the expressive ca-pabilities, by allowing to represent richer semantic re-lationships. Given the inherently open nature of per-vasive, sensor-driven systems, where the need to inte-grate domain knowledge and meaningfully aggregateand interpret low-level context information is a cru-cial requirement, it comes as no surprise that SemanticWeb technologies have been acknowledged as afford-ing a number of highly desirable features.

4. Ontology-modelled Pervasive Systems

This section provides an overview of how ontolo-gies are employed in a pervasive system by introduc-ing representative ontology models in the area. Gen-erally, there are two main functions of ontologies inpervasive computing: semantic enrichment at a con-ceptual level and support for developing knowledge-centric software [39]. The following examples are themost influential models and systems that demonstratethese two functions.

1http://edge.cs.drexel.edu/assemblies/software/owljesskb/

4.1. SOUPA

The Context Broker Architecture (CoBrA in Fig-ure 2), is a broker-centric, agent-based architecturesupporting context-aware computing in intelligentspaces [102]. CoBrA provides a means of acquir-ing, maintaining, and reasoning about context; shar-ing knowledge; detecting and resolving inconsistentknowledge; and protecting user privacy [35]. All of theabove capabilities are provided by the context broker,an intelligent agent that is the central component inCoBrA. Inside the context broker, the Context Acqui-sition Module offers a library of procedures that forma middleware abstraction for context acquisition; theContext Knowledge Base maintains a shared model ofcontext on the behalf of a community of agents and de-vices in the smart space; the Context Reasoning Enginereasons over the context so as to detect and resolve theinconsistent knowledge; and the Policy ManagementModule provides a set of user-defined inference rulesthat deduce instructions for deciding the right permis-sions for different computing entities to share a par-ticular piece of contextual information, and for select-ing the recipients to receive notifications of contextchanges.

Contextual information in CoBrA is represented bya set of ontologies, called COBRA-ONT that is im-plemented in OWL. COBRA-ONT is the key require-ment for modelling context in the smart meeting ap-plication [32]. It defines typical concepts and relationsfor describing physical locations, time, people, soft-ware agents, mobile devices and meeting events. Aset of more general ontologies, named SOUPA (Stan-dard Ontology for Ubiquitous and Pervasive Appli-cations), that encompasses COBRA-ONT, has beenproposed for supporting pervasive computing appli-cations. SOUPA [37] borrows terms from other stan-dard domain ontologies such as FOAF [65], DAML-Time [88], OpenCyc [111], RCC [23,46], and the ReiPolicy [101] Ontology.

SOUPA offers a formal, well-structured way tomodel context and thus provides rich semantics forprogramming. It also allows policies to be defined tosupport trust and privacy. This is demonstrated in Co-BrA’s EasyMeeting application [34], in which the on-tologies facilitate knowledge sharing and work withlogic inference rules to reason about context. SOUPAalso applies the MoGATU ontology to express the be-liefs, preferences, intentions, and desires of an agentor a user, which makes it possible to rank the prioritiesof plans and goals.


Information Servers(Exchange Server, iCal,

YahooGroups, etc)

Semantic Web &Web Services

(RDF, DAML+OIL & OWL)

ContextBrokerContext

Knowledge baseContext

Reasoning EngineContext

Acquisition ModulePrivacy

Management Module

Database(MySQL)

MeetingAgent

PersonalAgent

LunchAgent

Presentation Rec AgentEmergency

AlertAgent

SOAP + RDF/OWL RIPA-ACL + RDF/OWLContext-Aware Devices

Smart Tag Sensors(RFID)

Environment Sensors(Xanboo & X10 technology)

Device & Gadget Sensors(Java Ring, SmartCart etc)

Contexts in External Sources

Context-Aware Agents

Fig. 2. CoBrA – an intelligent context broker (source: [38])

4.2. Gaia

Gaia is an infrastructure for smart spaces, whichare pervasive computing environments that encompassphysical spaces [161]. The main characteristic of Gaiais that it brings the functionality of an operating systemto physical spaces. It employs common operation sys-tem functions (including events, signals, file systems,security, and processes), and extends them with con-text, location awareness, mobile computing devicesand actuators. Using this functionality, Gaia integratesthe devices and physical spaces, and allows the physi-cal and virtual entities to seamlessly interact.

Ontologies are introduced in Gaia as an efficientway to manage the diversity and complexity of de-scribing the resources (e.g., devices and services). Firstof all, they work as a system specification for configu-ration management by providing a standard machine-readable taxonomy of the different kinds of entities, in-cluding applications, services, devices, users, and datasources. Therefore, these ontologies are beneficial forsemantic discovery [120], matchmaking [199], inter-operability between entities [122], and interaction be-tween human users and computers [157].

Gaia ontologies [156] model contextual informa-tion including physical, environmental, personal, so-cial, application, and system contexts. They describerelations between different entities and establish ax-

ioms and constraints on the properties of the entitiesthat must be satisfied. However, Gaia does not definespecific ontologies for each of the core concepts of per-vasive computing. For example, the location context isrepresented in predicate form, where the location pred-icate must have a subject which belongs to the set ofpersons or things, a verb or preposition like ‘inside’or ‘entering’, and a location, which may be a room ora building, e.g., Location(chris, entering,

room 3231).In Gaia, the Ontology Server is responsible for load-

ing and validating ontologies from DAML+OIL docu-ments, and composing ontologies into a combined on-tology for the entire system. It is also capable of serv-ing logical queries to the Knowledge Base that repre-sents the dynamically integrated ontology base. TheOntology Explorer is a graphical user interface (GUI)that allows users to browse and search ontologies, andto interact with other entities in the space [121]. Theontology explorer offers a clearer understanding of thesemantics of a system. This also makes it easier towrite rules that determine context-sensitive behaviour.The explorer helps human users to interact directlywith the constituent parts of a pervasive computing en-vironment.


4.3. ASC

Aspect-Scale-Context (ASC) is a contextually basedmodel [186]. A context is a set of contextual informa-tion characterising entities (like a person, place, or ageneral object) relevant for a specific task in their rel-evant aspects. An aspect is a classification, symbol- orvalue-range, whose subsets are a super-set of all reach-able states, grouped in one or more related dimensionscalled scales [187]. The ASC model shows how con-textual information is used to characterise a state of anentity under a specific aspect.

Ontologies implemented in ASC serve as the un-derlying basis for describing contextual facts and in-terrelationships, which improve the formality of themodel. They also facilitate service discovery and ser-vice interoperability on the context level. CoOL, theContext Onotology Language [187] is derived fromASC to facilitate ontology-based contextual interoper-ability. CoOL is divided into two subsets: the CoOLCore, which projects the ASC model into common on-tology languages such as OWL and DAML+OIL, andF-Logic [104]; and CoOL Integration, which is a col-lection of schema and protocol extensions as well ascommon sub-concepts of ASC. OntoBroker [56], theinference engine, deals with knowledge query issuesby specifying knowledge as conclusions from ontolo-gies. The advantage of CoOL is that it enables contextinteroperability and context-awareness during servicediscovery and execution.

[187] describes an overall architecture of the sys-tem that focuses on the context provider domain. Themain parts of this middleware component include on-tologies, facts, rules, and the inference engine. The firsttwo are CoOL based knowledge, and with given rulesthey deliver knowledge to the inference engine. CoOLis capable of deriving new knowledge, validating con-sistency and asserting inter-ontology relationships soas to integrate ontologies.

4.4. SOCAM

The Service-Oriented Context-Aware Middleware(SOCAM in Figure 3) is an architecture which enablesthe building and rapid prototyping of context-awareservices in pervasive computing environments [80].The middleware abstracts physical spaces from whichcontexts are acquired into a semantic space where con-texts can be easily shared and accessed by context-aware services.

Context-aware Service Context-aware Service

ContextReasoner

ContextKB

ContextDatabase

Internal Context Provider

External Context Provider

Service LocatingService

Virtual Sensors(Web Sensors,

Information Servers)Physical Sensors

Con

text

Inte

rpre

ter

Context Application

Layer

Context Middleware

Layer

Context Sensing

Layer

Fig. 3. Architecture of SOCAM (source: [79])

The CONtext ONtology (CONON) is an ontology-based context model, in which a hierarchical approachis adopted for designing context ontologies [80]. Theontologies include a common upper ontology for thegeneral concepts in pervasive computing (such asperson, location, computing entity, and activity) anddomain-specific ontologies that apply to different sub-domains like smart homes. The context model supportsmultiple semantic contextual representations like con-textual classification, contextual dependency and qual-ity of context. Using CONON, two types of contex-tual reasoning tasks are supported: ontology reason-ing with DLs, and user-defined reasoning by definingspecific rules in first-order logic.

The CONON ontologies help sharing common un-derstanding of the structure of contextual informationfrom users, devices and services so as to support se-mantic interoperability and reuse of domain knowl-edge. They also support efficient reasoning mecha-nisms so as to check the consistency of context and de-duce higher-level, implicit context from raw context.A context-aware home scenario is implemented in theprototype system to demonstrate the work of ontolo-gies [80].

4.5. Ontonym

Ontonym [177] comprises a set of upper ontolo-gies [12] and a methodology for constructing high-level pervasive environment descriptions. Rather thansupport the complete specification of a pervasive sys-tem, Ontonym aims to provide a basis for semanticinteroperability across pervasive applications and en-vironment by providing a core set of concepts fromwhich application-level concepts can be defined bymeans of extension and composition.


Beyond commonly recurring concepts such as peo-ple, location, events, and resources, the core offeringsof Ontonym include: a model of time that supports thespecification of historical and predicted states; a sens-ing model that captures properties of data sources suchas their operational tolerances, capability, and sens-ing frequency; a model of provenance for recordingsources of data and the role of any intermediate com-ponents in its derivation [27]; and support for captur-ing imprecise information by means of an uncertaintyvocabulary.

Ontonym has a number of strengths. In particular, itis designed with orthogonality in mind, with the con-cepts in any two given ontologies loosely coupled inorder to support reuse. Particular attention is paid toreusing and linking to terms from existing standard vo-cabularies (e.g., OWL-Time [141] and the Measure-ment Units Vocabulary2), avoiding encoding bias; thatis, information is not represented for the convenienceof particular applications. Ontonym’s Person and Lo-cation vocabularies are more expressive than the offer-ings of other pervasive computing vocabularies, and itis the only pervasive computing vocabulary to considerthe representation of sensors and (basic) provenanceinformation.

4.6. Semantic Smart Home

Semantic Smart Home (SSH) is considered as an ex-tension of the current smart home solutions with se-mantic dimension, which will enable semantic-basedknowledge discovery and intelligent processing. Theessence of a SSH is that data, devices, and services aregiven well-defined meaning so that they can be linkedwithin and across smart homes. The semantic exten-sion makes it possible to exploit advanced semanticor artificial intelligence based information processingtechniques to provide value-added data processing ca-pabilities such as data integration, interoperability andhigh-level decision support [39].

SSH is realised in a layered conceptual architecture,consisting of the physical layer that hosts a numberof sensors or devices, the data layer that collects rawdata from sensors, the application layer that providesassisted living applications, and the semantic layer be-tween the data and application layers. The semanticlayer aims to provide a homogeneous view over het-erogeneous data, thus enabling seamless data access,

2http://purl.oclc.org/NET/muo/muo

Reusable Services

Third-Party Apps & Tools

Advanced Interaction Monitoring & Directing

Application Layer

Intelligent Processing Layer

Aggreg

ation

Learn

ing

Mining

Portal

Visualis

ation

Proven

ance

RDF Data Bus (RDF+OWL+HTTP+SPARQL)

SH1, SH2 in institution 1 SH1, SH2 in institution 2

Semantic Layer

Data LayerData Flat Files Database 1

Actuators Sensor 1 Sensor 2 MultimediaOutputsPhysical Layer

Fig. 4. The conceptual architecture of the SSH (source: [39])

sharing, integration and fusion across multiple organ-isations. Underlying the semantic layer are the smarthome ontologies for representing and reasoning ondata, including low-level sensor data, domain knowl-edge, and high-level human activities.

Among the smart home ontologies, there is an activ-ity ontology that models human activities as a hierar-chy of classes with each class described by a numberof properties. The engineered activity models capturebuilt-in interrelations between an activity and objectsthat are used when a user is performing the activity.For example, the activity MakeTea is a subclass ofMakeHotDrink, and hasDrinkType is a propertyof MakeHotDrink, which links the class Drink-Type (e.g., its instances are tea, coffee, or chocolate)to MakeHotDrink. The process of activity recogni-tion is to perform equivalency and subsumption rea-soning to check whether a conceptual description of anactivity formed by collected sensor data is equivalentto any atomic activity concept [41].

4.7. Top-level Ontologies in Smart HomeEnvironments

Most ontology models in the above systems followthe philosophical nature of ontology; that is, defininga commonly agreed vocabulary for representing dataand knowledge structure. The vocabulary makes data


understandable, sharable, and reusable by both humansand machines. However, developing such ontologiesusually takes a large amount of knowledge engineeringeffort from developers.

Ye et al. design and develop a top-level ontologymodel [219], which captures essential semantics com-mon to instances of all dimensions of an informa-tion space; i.e., levels of granularity, conflicting, andoverlapping relationships. For example in Temperaturedimension, an instance hot is defined on the range[30,∞], and another instance cold on [−∞, 0]. Wecan infer that hot is conceptually conflicting withcold. Meta-properties are explored on these three re-lationships, which facilitate deriving new knowledgefrom a limited amount of specified knowledge. Fromthe individual instances, context predicates and activ-ities are constructed. A context predicate is defined asa two-dimensional product space of information, andan activity is a multiple-dimensional product space ofinformation that indicates a situation occurring in theenvironment. Generic reasoning rules are specified toderive relationships between context predicates and re-lationships between activities. The derived relation-ships facilitate system-level tasks in pervasive comput-ing, including detecting contradictory sensor events,checking consistency of activity rules, and aiding ap-plication design on an inferred hierarchy of activities.

Schmidtke et al [170] also propose a top-level on-tology, which is built on the theory of mereotopology.Mereotopology is a formal theory of part and wholethat has been widely used in the area of formal on-tology. From a primitive relation part-hood, they dis-cuss the overlap, underlap, and connection relationsbetween contexts. Two contexts overlap if and only ifthey share a part that is not identical to the empty (i.e.,non-existing) context, while two contexts underlap ifand only if they are both part of a context that is notidentical to the whole domain. A connection can beintroduced between two contexts, which is a reflexiveand symmetric relation. These relations are consideredcrucial in modelling the notions of reachability in aconceptual hierarchy.

4.8. Summary

Table 3 shows that of the requirements introduced inSection 2.2.1,the above ontology models mainly focuson conceptual modelling and reasoning, while the ser-vice discovery and privacy and trust are added valuesto the semantic modelling. The following sections in-

troduce additional systems and ontologies beyond theabove approaches.

In terms of the Semantic Web technologies intro-duced in Section 3, OWL DL is the language used mostby the afore-described systems. They are usually con-figured through multiple reasoners, among which themost used ontological reasoners are Racer, FaCT andPellet. As yet, OWL 2 profiles have not been notablyemployed.

5. Integrating Semantic Web with PervasiveSystems

The aforementioned outlined key Semantic Webfeatures and provided a first overview of ontology-based pervasive systems. In the following we explorein detail how these features meet the requirements oftypical pervasive systems, including the representationsensor data, the modelling and querying of contextand events, context-driven reasoning, addressing un-certainty, semantic service discovery, privacy and trust.

5.1. Representing Sensor Data

Hossein et al. propose a Sensory Data Set Descrip-tion Language (SDDL), which is an XML-encodeddescription language for sensor data originating frompervasive spaces [94]. The scope of SDDL is to specifyinformation about the pervasive space, including: (1)available sensors/actuators such as the sensor id, mea-surement unit, its type, and where it is installed, (2)data set parameters such as the space layout of the en-vironment, the project specification, and contact infor-mation, and (3) sensor events such as the timestamp,activated sensor id and its value, and whether or not itis noisy. It does not concern physical properties of thesensors. The motivation behind this work is to providea standard description language for datasets so as tofacilitate sharing the existing and future datasets.

Built on SDDL, Ye et al. propose a generic con-ceptual model PI to model sensors and their qualitymeasure for existing smart home data sets [220]. Eachsensor type has a set of properties including its name,id, domain – a parameter that describes what the sen-sor type measures (e.g., acceleration or usage of gas),and a unitOfMeasurement – the unit used to report ob-served values. Each sensor type can be associated witha technical specification that includes its manufacturer,model, size, deviation of readings produced by its sen-sors (e.g., ‘a maximum precision of ± 1.5% full scale’


Table 3Evaluation of requirements in pervasive computing satisfied byontology-based models in Section 4

Ontology-based Systems Conceptual modelling Reasoning Service discovery Privacy and trustCoBrA (SOUPA)

√ √–

√

Gaia√ √ √ √

ASC (CoOL)√ √ √

–SOCAM (CONON)

√ √ √–

Ontonym√ √

– –Semantic smart home

√ √– –

Top-level ontologies√ √

– –

idnameunitOfMeasurementdomain

SensorType modelsizevalueRangemanufacturerfrequencydeviation

TechSpecification

0..11

idnametypeIdobject

Sensor

idlocationfunction

Object

*1

1 11 1..nsensorIdvalueassertedAtvalidDuring

Observation

Fig. 5. The sensor and quality specification model in PI (source:[220])

for a gas flow sensor), its sampling frequency, anda number of valueRange parameters – boundary val-ues that characterise different states. These values maycome from the manufacturer or through the applica-tion of learning techniques. It supports deriving high-level context from raw sensor readings; e.g., a contextstoveOn is inferred if the gas sensor on the stove hasa reading greater than a threshold.

[stoveONRule:(?A sensor:domain "GAS"),(?A sensor:value ?B),(?A sensor:object ?C),(?C rdf:type :Stove)ge(?B "1"^^xsd:double)

-> createClassification(?A, "stoveOn")]

SensorML, the Sensor Model Language, is specifiedas a generic data model in UML for capturing classesand associations that are common to all sensors. It ispart of an Open Geospatial Consortium (OGC) ini-tiative to contribute to the development of a SensorWeb [165,164]. SensorML provides an XML encodingschema for developing a persistent data store for sensormetadata and sensed attribute values. Chu et al. applyit to model web-hosted sensors to enable remote dis-covery, access and usage of real-time data [44]. Sen-

sorML is a syntactic model for sensors, while the se-mantic issues such as interpreting and integrating itsencoded information have not been fully resolved [48].

Based on SensorML, Russomanno et al. propose acomprehensive deep sensor ontology [164], called On-toSensor. OntoSensor also consists of other standardi-sation such as the IEEE Suggested Upper Merged On-tology (SUMO) [189] and ISO 19115 [96]. OntoSen-sor includes a domain theory expressed in a languagethat is constructed using the functional and relationalbasis sets to support ontology-driven inference. It isable to capture sensors’ computing and communica-tion capabilities and their percept attributes that areused to store measurements of physical phenomena,and detection, classification and tracking of physicalobjects. For example, a thermal sensor is associatedwith an attribute sample_interval on which aprocessing service detectAlarm(threshold) isdefined; that is, if the thermal expansion or contractionat regular sample intervals are beyond a threshold, thenan alarm is reported. OntoSensor also includes moreadvanced inference mechanisms that can be used forsynergistic fusion of heterogeneous data.

More sensor ontologies can be found in a reviewdone by the W3C [209] and Compton et al [48].Based on the review, the W3C Semantic Sensor Net-work Incubator group [208] has developed the Seman-tic Sensor Network (SSN) ontology [47]. It targets atthe formal and machine-processable representation ofsensor capabilities, properties, observations and mea-surement processes, with which it aids in searching,querying, and managing the sensor network and itsdata. Central to the ontology is the Stimulus-Sensor-Observation (SSO) ontology design pattern [98] thatprovides a lightweight model for representing sensors,their inputs (called stimulus) and observations. SSOis reusable for a variety of application areas and itcan be used in conjunction with other relevant ontolo-


gies. Both SSN and SSO have been aligned with theDOLCE+DnS Ultralite (DUL) ontology [64] so as toenable the integration into more complex ontologiesas a common ground for alignment, matching, transla-tion, or interoperability.

5.1.1. AnalysisThe above sensor-centric and observation-centric

ontologies have developed from a syntactic model to asemantic and knowledge model. SDDL andPI tend toshare and reuse sensor data sets by describing sensorsand data in a light-weight syntactic model. PI sup-ports classifying raw sensor data into high-level con-text through user-specified rules. SDDL and PI focuson data, while more standard solutions like SensorMLand SSN aim to standardise interfaces for services anddescription languages for sensors and their processesto enable syntactic interoperation [48]. Sensor ontolo-gies are moving towards a deeper knowledge modellike OntoSensor and SWAMO [213] to express and au-tomatically create a composition of processes of sen-sors [48]. To do this, the current sensor ontologies shallwork towards enabling a variety of uncertainty rea-soning mechanisms (e.g., fuzzy and probabilistic rea-soning) and more advanced technologies like mixturesof DL, structural similarity and information retrievaltechniques [48].

5.2. Modelling and Querying Context

The nature of pervasive computing is such thatthe applications encompassed by this broad terms arevastly heterogeneous in nature and broad in their datarequirements and scope. Hence, in an open environ-ment it is impossible to define, without a priori knowl-edge of the applications that will use it, a ‘complete’model of content. However we may straightforwardlyobserve that a large number of applications exhibitoverlapping data requirements, the most common ofwhich are the need to represent time, location, actors,and resources. A vocabulary for time supports the tem-poral localisation of entities; in supporting spatial lo-calisation, location has multiple representations, eachsuited to particular classes of application; people andagents are central actors in pervasive applications; and,the notion of a resource encompasses myriad things,both physical and virtual with which a user or applica-tion may interact. These neatly correspond to the no-tions of when, where, who, and what introduced in sec-tion 2.3.

5.2.1. When—TimeTemporal features play a key role in enhancing en-

tity descriptions within a pervasive environment. Forexample, representing the time at which an event takesplace, the frequency with which a sensor samples theenvironment (e.g., every 10 seconds), or the expectedduration of an activity (e.g., 30 minutes). Here, wemust also consider semantically meaningful notions oftime, which may have an exact (e.g., Monday morn-ing, Christmas) or fuzzy (e.g., lunch time, car journeyduration) correspondence to physical time.

The most common mechanical representation forphysical time is the ISO 8601 standard [95], whichis based on the Gregorian calendar, and provides alexical format for modelling dates, times, durations,time intervals, and time zones. For example, 2012-05-10T09:32BST represents the time of 9:32am onthe 10th of May, 2012 in British Summer Time. Asubset of the lexical formats defined by this standardare adopted by XML schema, which RDF adopts as ameans of typing data literals.

Given any two temporal features, there exists a rela-tionship between them that we may also wish to model.Instants are related to intervals by the notion of con-tainment, while Allen’s temporal calculus [4] definesthe following seven relationships between time inter-vals

– during(t1, t2): time interval t1 is fully containedwithin t2;

– starts(t1, t2): time interval t1 shares the same be-ginning as t2, but ends before t2 ends;

– finishes(t1, t2): time interval t1 shares the sameend as t2, but begins after t2 begins;

– before(t1, t2): time interval t1 is before intervalt2, and they do not overlap in any way;

– overlap(t1, t2): interval t1 starts before t2, andthey overlap;

– meets(t1, t2): interval t1 is before interval t2, butthere is no interval between them, i.e., t1 endswhere t2 starts;

– equal(t1, t2): t1 and t2 are the same interval.

and their inverses (contains, startedBy, finishedBy, af-ter, overlappedBy, and metBy) for a total of thirteenrelationships (equals being its own inverse).

The W3C OWL-Time ontology [141], which pro-vides a vocabulary for Allen’s thirteen temporal rela-tions, uses XML Schema’s dateTime formats, butalso provides its own component based on Date-TimeDescription that can express additional in-formation (e.g., ‘day of week’ and ‘day of year’) to


make mechanical extraction of and reasoning on thisinformation easier. However, we note that as a tem-poral relationship is formed by each pair of temporalentities, it is impractical to manually specify or con-cretely realise these in any data model of notable size.This indicates a need for special consideration whenevaluating these temporal predicates as part of a query.

Of the ontology modelled systems described in theprevious section, SOUPA’s temporal predicates arebased on DAML-Time (OWL-Time’s predecessor),while Ontonym adopts OWL-Time directly in its mod-elling of temporal concepts. Temporal concepts maybe straightforwardly mapped to the top level ontolo-gies discussed. Although it is possible to represent allthe temporal relations, only a subset of the relationsmap to the properties of the underlying model (e.g.,during or overlap) upon which standard inference isperformed.

5.2.2. Where—LocationLocation information provides a means of invoking

application behaviour based on the real world position-ing of users and artefacts. Location information has anumber of possible representations, ranging from ab-solute, to relative, to symbolic, all of which can be re-lated, and with specific forms more appropriate thanothers for any given application [60]. Location is oftenregarded as the most important type of context, and lo-cation models and query frameworks had been exten-sively researched, before the emergence of the Seman-tic Web (e.g., [89,87,100,155]).

The CONON location ontologies exist primarily todifferentiate between indoor and outdoor spaces, whileASC’s location ontologies have limited scope, support-ing the representation of spatial positions in either theWGS84 and Gauss-Kreuger coordinate systems, andrepresenting the distances between positions.

More comprehensive is SOUPA’s location ontology,which is formed through the conjunction of two ex-isting location vocabularies, OpenCyc [111] and Re-gion Connection Calculus (RCC) [23,46]. The Open-Cyc vocabulary supports the symbolic representationof spaces, while RCC provides a set of spatial rela-tions (e.g., overlap, disconnection, tangental part) thatmay hold between two regions—essentially the loca-tion equivalent of Allen’s temporal relations discussedabove. Ontonym [177] adopts a simpler spatial re-lation model, supporting only containment, overlap,and adjacency. It supports multiple coordinate systems(based on the coordinate translation scheme of Jianget al. [100]) and incorporates a notion of relative posi-

tioning between entities based on compass directionsand distance. The LOC8 framework uses this modelto support conversion of positioning data between co-ordinates, symbolic locations and relative positions. Italso provides an API for querying for the position ofentities, returning a list of entities within a spatial re-gion, and generating paths between two points in themodel [179].

5.2.3. Who—Person/AgentThis context type broadly relates to the actors in a

pervasive system, which are typically people or agentswhose reputations are manifest by software. The datarepresentation requirements associated with these enti-ties are usually application (or a class of applications)specific, and focus on the role played, or interactionsexpected of an entity in a scenario.

The CONON person ontology is tightly-coupled toits application offerings and provides a small vocab-ulary for describing a person’s name, age, and situa-tion. SOUPA builds upon the FOAF [65] to support theexpression of human profile information (name, age,contact details), and the MoGATU BDI ontology [147]to support the description of agent state – their beliefs,desires, and intensions – the detail of which is an appli-cation specific concern. Ontonym’s person vocabularyis based on a subset of terms from vCard3, W3C PIM4

and FOAF, and provides support for modelling date ofbirth, gender, language, and contact profiles, postal andemail addresses, telephone and fax numbers, and webpresence. Ontonym provides a component based namemodel that supports the semantic modelling of nameterms (e.g., ProfessionalTitle, GivenName,and PatronymicName), and borrows from Daviesand Vitiello’s relationship vocabulary for FOAF [71]to define relations between people covering genetic,working, romantic, residential, and friendship connec-tions.

5.2.4. What—ResourceGiven that the term resource is itself generic, the

lack of an all encompassing vocabulary for describingresources is not surprising. A resource may refer toanything not otherwise explicitly modelled in a partic-ular pervasive system, for example, a device, an image,a document, a physical object without computationalpower, or a virtual artefact. While, for the most part,the properties of a resource one might wish to modelare application specific in nature, resource descriptions

3vCard ontology: http://www.w3.org/2006/vcard4W3C PIM: http://www.w3.org/2000/10/swap/pim/


might draw from multiple ontologies. For example,FOAF provides some general terms for describing im-ages, documents, and projects, the Dublin Core Meta-data Initiative vocabulary5 provides terms for describ-ing aspects of resources such as their creators, repre-sentation format, and licensing, and the GoodRelationsontology [85] provides an e-commerce vocabulary fordescribing products and services.

5.2.5. AnalysisThis section has evaluated the way in which the

most common elements of primary context in perva-sive computing are represented. While one can feasi-bly conceive of the widespread adoption of a single,shared temporal model and location model upon whichevent descriptions are based, the specification of theactors and resources in a pervasive environment willremain highly application specific. Where the need tomodel particular features of such entities occurs sep-arately (i.e., using separate vocabularies), the promiseof Linked Data [19] plays a critical role in support-ing the integration of entities with diverse descriptionsacross applications and environments.

5.3. Event-centric modelling and reasoning

The efficient representation and processing of eventsis an important and challenging task in pervasive en-vironments. In most cases, only a small number or acombination of raw primitive events is of real interest.Such low-level events are produced directly by the sen-sors or after low-level data processing. However, high-level, real-world complex events, such as situationsand human activities in Figure 1, cannot be recognisedor processed, due to the absence of an expressive do-main knowledge that would enable the further corre-lation of events and multi-modal information beyondpredefined patterns and attributes. Moreover, in het-erogeneous environments with a large volume of sen-sors where data is gathered from heterogeneous sys-tems with potentially different terminologies, the rep-resentation, integration and correlation of event knowl-edge raise new challenges.

In order to overcome the above limitations, researchefforts have focused on the definition of ontology-based event models. Ontologies are used in this con-text as common vocabularies for representing knowl-edge relevant to events and their relationships, solv-

5DCMI terms: http://dublincore.org/documents/dcmi-terms/

ing interoperability problems and providing the meansfor high-level event interpretations based on ontol-ogy reasoners. This section reviews existing domain-independent ontologies for event modelling, present-ing the different design patterns and expressive capa-bilities they provide. In addition, research efforts to-wards the development of semantic complex event pro-cessing (SCEP) frameworks are described that enrichthe results of traditional CEP engines with ontology-based background knowledge and reasoning.

5.3.1. SOUPAEvents in SOUPA are modelled as instances of the

soupa:Event class that represents a set of all eventsin the domain. However, this class is time and spaceagnostic in the sense that it does not provide proper-ties for directly associating events with temporal andspatial information.

Spatiotemporal information in SOUPA can be mod-elled in terms of the soupa:SpatialTemporal-Thing class that is defined as the intersection of thesoupa:TemporalThing and soupa:Spatial-Thing classes. Particularly for events, the soupa:-SpatialTemporalEvent class is defined as theintersection of the soupa:Event and soupa:Spa-tialTemporalThing.

5.3.2. OntonymThe event ontology of Ontonym provides a means

of describing activities of interest within a system. Theontonym:Event class is defined as the union of theontonym:InstantEvent and ontonym:Inte-rvalEvent classes that are used to model eventsthat occur instantaneously or over a time period, re-spectively. Spatial information on events can be ex-pressed through the ontonym:SpatioInstant-Event and ontonym:SpatioIntervalEventclasses. Any temporal information in Ontonym is ex-pressed using the ontology constructs of the OWL-Time ontology. Spaces are represented using a combi-nation of ontonym:SymbolicRepresentation,ontonym:GeometricRegion and ontonym:-RelativeLocation ontology classes.

A role played by a person, device, resource, or ob-ject in an event is represented as an instance of theontonym:Role class. The ontonym:contains-Role property identifies the types of roles that canbe taken in an event and the ontonym:playsRoleproperty is used to associated an entity with an event.


sem:hasSubEvent

sem:Event sem:Actor sem:Place sem:Time

sem:hasTime

sem:hasActor

sem:hasPlace

sem:PlaceType

sem:placeType

sem:EventType

sem:eventType

sem:ActorType

sem:actorType

sem:TimeType

sem:Type

sem:timeType

sem:Core

sem:Rolesem:Temporary

sem:Constraintsem:View

sem:RoleType

sem:roleType

sem:hasTimeStamp

sem:hasSubType

Core Classes(Foreign)

Type SystemProperty

Constraints

the Simple Event Model (SEM)

Literal sem:hasTimeStamp

Literal sem:hasTimeStamp

sem:accordingTosem:Authority

sem:hasTime

Willem Robert van HageVéronique MalaiséVrije Universiteit Amsterdam

Fig. 6. The classes of the Simple Event Model ontology [203]. Arrows with open arrow heads symbolise rdfs:subClassOf properties andregular arrows visualise rdfs:domain and rdfs:range restrictions

5.3.3. The Event OntologyThe Event Ontology (EO)6 is a simple upper-level

ontology for modelling events, defining three classes(eo:Event, eo:Product and eo:Factor) andeighteen properties. An event may have sub-events(eo:sub_event), actively participating agents (eo:-agent), passive factors (eo:factor), products(eo:product, i.e. everything produced by an event),and a location in space (eo:place) and time (eo:-time) by reusing the WGS84 Geo Positioning7 andOWL-Time ontologies, respectively.

The Event Ontology has been mainly used for defin-ing and interlinking events in the LOD cloud [173] andfor annotating and extracting multimedia events, suchas recordings and compositions, being part of the Mu-sic Ontology8. For example, an extension of the MusicOntology has been used in [175] in order to develop anontology-based music recommendation system basedon user’s emotions/moods that combines user profilesand event (situation) reasoning.

6http://motools.sourceforge.net/event/event.html

7http://www.w3.org/2003/01/geo/wgs84_pos8http://www.musicontology.com

5.3.4. Simple Event ModelThe Simple Event Model (SEM) [203] is an effort

to define an ontology model for events without strongsemantic constraints. This decision is justified by theopen nature of the Web and the need to model different

(even conflicting) views of the same event. The lack,however, of strong semantic constraints, such as func-tional properties, disjoint classes and cardinality re-strictions, hampers the ability to automatically validateand resolve model inconsistencies using formal infer-ence mechanisms. Therefore, SEM is characterised bya tradeoff between model reusability and automatedreasoning and validation capabilities. Figure 6 presentsthe relationships among the classes of the SEM ontol-ogy.

SEM defines four core classes: sem:Event (formodelling events), sem:Actor (who or what partici-pated in an event), sem:Place and sem:Time (cap-turing where and when the event takes place). Eachcore class is associated with the sem:Type class thatis used to aggregate implementations of type systemsfrom other ontologies, such as the Getty Thesaurusof Geographic Names9. Therefore, the instance typesin SEM can be either instances or classes from for-

9http://www.getty.edu/research/tools/vocabularies/tgn/index.html


eign vocabularies, targeting at the reusability of exist-ing type vocabularies, regardless of how the types areimplemented.

There are two main property types in SEM: sem:-EventProperty and sem:type. The former isused to correlate instances of the sem:Event classwith other instances of arbitrary classes and the lat-ter correlates instances of the sem:Core class withinstances of the sem:Type class. There are alsotwo aggregation relations. The sem:hasSubEventcan be used to define specialisation of events. Sim-ilarly, the sem:hasSubType relates instances ofthe sem:Type class. Finally, there are seven time-related properties: sem:hasTimeStamp (for sin-gle time values), sem:hasBeginTimeStamp andsem:hasEndTimeStamp (for time intervals), sem:hasEarliestBeginTimeStamp, sem:hasLatestBeginTimeStamp, sem:hasEarliestEndTimeStamp and sem:hasLatestEndTimeStamp(for uncertain time intervals).

Property constraints in SEM can be defined ei-ther as a reification of the property or by turning theproperty into n-ary relation. The classes sem:Role,sem:Temporary and sem:View are three types ofSEM constraints for defining the role of a class in-stance (e.g., sem:Actor) in an event, the temporalboundary within which a property holds and the dif-ferent points of view, respectively. More expressive re-lationships among events, such as causality [168], oractive/passive actor participation semantics cannot bemodelled in SEM.

SEM defines also mappings of the model on otherontologies, e.g., LODE [173], based on the SKOS vo-cabulary. Furthermore, the developers provide a Pro-log API for SEM in order to ease the procedure of pop-ulating the ontology with instances.

5.3.5. Event-Model-FEvent-Model-F [168] defines an expressive model

for capturing and representing occurrences in the realworld. It is based on DUL, following the descriptionsand situations ontology design pattern (DnS) [73] formodelling aspects of events, such as object partici-pation, mereological, causal, and correlative relation-ships, and different interpretations of the same event(by reifying events in order to describe n-ary relations).The Event-Model-F ontology introduces six ontologydesign patterns that are described briefly in the rest ofthis section.

Participation Pattern. This pattern is used to modelthe participation of objects (dul:Object) in events(dul:Event). The participation is expressed by aninstance of the F:EventParticipationSituat-ion class that satisfies an instance of the F:Eve-ntParticipationDescription class. The sit-uation includes the event and the participating objects,while the description classifies the event (dul:EventT-ype) and the participants (dul:Role).

Mereology Pattern. Compositions of events are ex-pressed using the mereology pattern (Figure 7). Thecomposite event and the component events are clas-sified by the F:Composite and F:Componentclasses, respectively, and they are all referenced by asituation instance (F:EventCompositionSitu-ation). The situation instance satisfies a descriptioninstance (F:CompositionDescription) that clas-sifies the composite and component events. Further-more, events may be temporally related to each otherand such relations can be expressed using the DOLCEvocabulary. D5.2.1 - V1.0

Description

EventCorrelationDescription

EventCorrelationSituation

Situation

Justification

EventType

Concept

Event Description

classifies

defines

isEventIncludedIn

satisfies

isObjectIncludedIn

classifies

Correlate

Correlate

Role

Figure 7: Correlation Pattern

ships. Thus, using a single causality relationship might not be precise enough. Consequently,her logic for causality allows for explicitly modeling the different kinds of causal relationshipsfound in literature. This can be supported with our Event-Model-F by specializing the causalitypattern to specific kinds of causal relationships.

3.4.4 Correlation Pattern

A set of events is called correlated if they have a common cause. However, there exists nocausal relationship between the two events [57]. The common cause may originate from asingle or a chain of multiple preceding cause-effect relationships. Correlation also differs fromco-occurrence where two or more events just (randomly) happen at the same time and do nothave a common cause.

Correlation is not of metaphysical interest as it is a property that can be derived from causality,i.e., the common cause. In our ontology, we model correlation explicitly as in many casesthe (correlating) effects of some common cause may be known, while the cause itself is not.The correlation pattern depicted in Figure 7 defines the role F:Correlate to classify theevents that are correlated. The Justification role classifies some Description, whichexplains the correlation in terms of a (mathematical) law or some theory.

3.4.5 Documentation Pattern

Documentary evidence for an event may be given by arbitrary objects, e.g., some sensor dataor media data, or by other events. Formally, this relation is expressed by the documentationpattern depicted in Figure 8. It defines the concept F:DocumentedEvent that classifies thedocumented event and the concept F:Documenter that classifies the documentary evidencefor that event. This evidence can be expressed by any specialization of an Object, e.g., a dig-ital photo taken with a cell phone during an incident, or a specialization of Event. Thus, thedocumentation pattern enables the inclusion of evidences described in ontologies other than F.For example, digital media data like images and videos can be classified as Documenter andprecisely described using, e.g., the Core Ontology for Multimedia [2, 60]. Thus, the documena-tion pattern connects the representation of the real-world events with the media assets that weretaken during these events. The objects are documented via the events in which they participate(see participation pattern in Section 3.4.1).

Page 28

Fig. 8. The correlation pattern in Event-Model-F (source: [169])

Causality Pattern. In Event-Model-F, causes and ef-fects are events and the causality pattern is used to ex-press relationships between events that play the rolesof causes and effects. The F:Cause and F:Effectclasses are used to classify events as causes and ef-fects, respectively, and the F:Justification classis used to classify a dul:Description instanceunder the justification of some theory.

Correlation Pattern. Two or more events are corre-lated if they have a common cause and there is nocause-effect relationship among them (causality). Thecorrelation pattern defines the role F:Correlatethat classifies correlated events. Similarly tothe causality pattern, the F:Justification roleclassifies a dul:Description instance thatexplains the correlation.


D5.2.1 - V1.0

3.4.2 Mereology Pattern

Events are commonly considered at different abstraction levels depending on the view andthe knowledge of a spectator. For instance, the event of a flooded cellar may be consid-ered as such or as part of the larger event of a flooding in which many other (smaller) in-cidents occur. The mereology pattern shown in Figure 5 enables expressing such mereolog-ical relations as composition of events. The composite event is the “whole” and the com-ponent events are its “parts”. Formally, a F:EventCompositionSituation includesone instance of an event that is classified by the concept F:Composite and many eventsclassified as its F:Component(s). Accordingly, an EventCompositionSituationsatisfies a F:CompositionDescription that defines the concepts Compositeand Component for classifying the composite event and its component events.

EventCompositionDescription

Description

Concept Parameter

EventType

Component

EventCompositionConstraint

TemporalConstraint

SpatioTemporalConstraint

Situation

Event

Quality

TimeIntervalsatisfies

defines

classifies

parametrizes

classifies

isEventIncludedIn

isTimeIncludedIn

isParameterFor

SpatialConstraint

SpatioTemporalRegion

SpaceRegion

parametrizes

parametrizes

Object

isSpaceTimeIncludedIn

isSpaceIncludedIn

hasParticipant

hasRegion

EventCompositionSituation

Composite

hasQualityhasQuality

defines

Figure 5: Mereology Pattern

Events that play the Component role may be further qualified by temporal, spatial, and spatio-temporal constraints. As events are formally defined as entities that exist in time and not in space(cf. Section 3.1), constraints including spatial restrictions are expressed through the objectsparticipating in the component event. For instance, a Component event may be required tooccur within a certain time-interval, e.g., the second week of June 2009. Depending on itsobjects, a Component event may also happen in a certain spatial region. For example, theflooding of a town should be composed of events that have objects associated to it, which havesome certain range of longitude and latitude. Finally, events and the objects bond to it maybe qualified by a spatio-temporal quality like the progress of a flood that extents over time andspace, starting with a high water level located in some area of a river and extending spatiallyover time into other areas. Any such constraints are formally expressed by one or multipleinstances of the F:EventCompositionConstraint. They define qualifying values fortemporal, spatial, and spatio-temporal qualities of a component event as shown in Figure 5.Thus, with the composition pattern, events may be arbitrarily temporally related to each other,i.e., they might be disjoint, overlapping, or otherwise ordered. In order to express such relativetemporal relations between events, one can facilitate the provided means of DOLCE such as theformalization of Allen’s Time Calculus8.8http://wiki.loa-cnr.it/index.php/LoaWiki:Ontologies

Page 26

Fig. 7. The mereology pattern in Event-Model-F (source: [169])

Documentation Pattern. This pattern is used to pro-vide documentary evidences for events. The docu-mented event is classified by the class F:Docume-ntedEvent and the documentary evidence is classi-fied by the class F:Documenter. The evidence maybe a specialisation of the Object class or anotherevent.

Interpretation Pattern. The perception of events heav-ily depends on the context and the viewpoint of theobserver and the explicit modelling of such contex-tual views is crucial. In Event-Model-F, such differentviewpoints can be expressed by instantiating accord-ingly the aforementioned designed patterns and bind-ing them with the interpretation pattern. The patterndefines the F:Interpretant class that classifiesthe event under interpretation. Each interpretation fur-ther classifies the relevant situations, that is, the par-ticipation, mereology, causality, correlation, and docu-mentation pattern instantiations.

5.3.6. Linking Open Descriptions of EventsThe Linking Open Descriptions of Events (LODE)

ontology [173] provides a model for representing andpublishing events that have happened in the past asLinked Data [19]. The model has been designed insuch a way, so as to allow the modelling of four ba-sic aspects of events: what, where, when, and who.The LODE vocabulary is aligned with other event-related vocabularies and ontologies, such as DUL,Event Ontology (EO) and CIDOC [61], and consistsof one class for representing events (lode:Event)

and seven properties. In contrast to Event-Model-F,LODE focuses on representing stable characteristicsof events, excluding parthood and causal relations thatbelong to the interpretative dimension of event mod-elling.

The lode:Event class is defined as subclassof the E2_Temporal_Entity class of CIDOCand equivalent to the eo:Event and dul:Eventclasses. In this way, an event in LODE representssomething that happened over a limited extent in timewithout specifying a change of state or attempting todistinguish events from processes or states.

In LODE, an event can be associated with at mostone interval of time through the lode:atTime prop-erty that is defined as subproperty of the CIDOC’sP4.has_time-span and dul:isObservableAtproperties. It is also equivalent to eo:time and thetime intervals are defined using the TemporalEn-tity class of the OWL-Time ontology.

The lode:inSpace property relates an eventto some spatial boundary (at most one), i.e. a re-gion of space, and it is defined as a subpropertyof dul:hasRegion, equivalent to eo:place andsuper-property of CIDOC’s P7.took_place-atproperty. Events can be also related to abstract placesthrough the lode:atPlace property that is sub-property of dul:hasLocation.

The association of events with objects and agentsis performed thought the lode:involved prop-erty (equivalent to DUL’s hasParticipant) and


its subproperty lode:involvedAgent (equivalentto DUL’s involvesAgent and super-property ofeo:agent), respectively.

5.3.7. Semantic Complex Event ProcessingState-of-the-art Complex Event Processing (CEP)

frameworks [163] are able to efficiently recognisedcomplex events based on predefined event-patterns,event-hierarchies or other event relationships (e.g.,temporal). CEP engines are capable of binding to in-put streams of real-time structured events that origi-nate either directly by sensors or after low-level featureprocessing. The data of the event streams are checkedfor specific values or correlations with data from otherstreams in order to identify and extract useful patternsthat are described using rules. The key feature is thesupport of temporal relationships (e.g., Allen’s tem-poral interval relationships) and aggregation operatorsthat enable the identification of complex correlationsamong the generated events.

Semantic Complex Event Processing (SCEP) is aneffort to improve the results of CEP by incorporatingontologies and rules into the process of complex eventdetection [128]. An abstract CEP architecture is pre-sented in Figure 9. Ontologies are used in this con-text as common vocabularies for representing knowl-edge relevant to events, such as event patterns, par-ticipants and their relationships, solving interoperabil-ity problems. Moreover, low-level events can be se-mantically associated with high-level domain conceptsof background ontological knowledge, improving thequality of event and activity recognition using contex-tual information. Usually, the semantic association ofthe events with background knowledge is further en-hanced with rules so as to overcome expressive lim-itations of the underlying ontology languages, as de-scribed in section 3.4. In this way, the native capa-bilities of CEP engines for real-time event processingcan be integrated into ontology-based pervasive frame-works for time-aware and high-level interpretation ofsensory data.

In [192], a commercial CEP engine (Coral810) iscombined with ontology-based reasoning. The plat-form aims at the semantic configuration of the CEPengine using domain ontologies about events, sensors,environments, etc. The event ontology extends SSNand it is used as a repository of complex event defini-tions that users define. Complex events are described

10http://www.aleri.com/products/aleri-cep/coral8-engine

4 Kia Teymourian, Malte Rohde, Ahmad Hasan, and Adrian Paschke

knowledge updates is not very high as the rate of the main event stream, e.g., frequencyof happening of music concerts compare to changes in a music band.

Fig. 1. High Level Architecture of Semantic-Enabled Complex Event Processing

4 Experiments and Demonstration

For our experiments, we use Prova4 as reaction rule language formalization and as arule-based execution which can be used as event processing engine. Prova uses reactivemessaging5, reaction groups and guards6 for complex event processing. Multiple mes-sages can be revived using revMult(XID, Protocol, Destination, Performative, Payload); XID, a conversation id of the message; Protocol, message passing protocol; Destina-tion, an endpoint; Performative, message type; Payload, the content of message. Provaimplements a new inference extension called literal guards. During the unification onlyif a guard condition evaluates to true, the target rule will proceed with further evaluation.We implemented the sparql_select built-in7 to run SPARQL queries from Prova whichcan start a SPARQL query from inside Prova on an RDF file or a SPARQL endpoint.This buit-in can use results which come from the SPARQL query and use them insideProva. It also provides the possibility to replace variables in SPARQL string which arestarting with $ with variables.

In our experiments we use the Prova rule engine. We use the sparql_select built-in:the rule engine first sends the embedded SPARQL query to triple store, gets the resultsback and then waits for incoming event stream to process. It processes the sequence of

4 Prova, ISO Prolog syntax with extensions http://prova.ws , July 20115 Prova Reactive Messaging http://www.prova.ws/confluence/display/RM/Reactive+messaging , July 2011

6 Event Processing Using Reaction Groups http://www.prova.ws/confluence/display/EP/Event+processing+using+reaction+groups, July 2011

7 Source codes for Semantic Web extensions in Prova 3 can be found in https://mandarax.svn.sourceforge.net/svnroot/mandarax/prova3/prova-compact/branches/prova3-sw/ , July 2011

Fig. 9. An abstract SCEP architecture (source: [143])

in terms of atomic events and the alerts that should betriggered upon detection. These events are transformedinto CEP streams and queries for the configuration ofthe CEP engine.

Teymourian and Paschke [195] present a frameworkfor semantic rule-based complex events processing.The ontology based representation of event data is per-formed by a map of the plain attribute-value pairs toa set of RDF triples (RDF graph) that can be used asevent instances. In this way, events are represented interms of URIs and they can be further interlinked withother ontology-based domain knowledge. More com-plex events can be retrieved by executing SPARQLqueries that match complex graph patterns. The reac-tion rule language of Prova11 is used in order to im-plement temporal reasoning operators and to performreasoning on the ontology-based knowledge.

ETALIS [6] is a CEP engine implemented in Pro-log that supports the definition and execution of EP-SPARQL queries, as well as the use of backgroundknowledge in a form of RDF ontologies [5]. EP-SPARQL is a language for event processing and streamreasoning that is used in order to detect events within astream of RDF triples (low-level events). EP-SPARQLextends SPARQL with a set of binary operators (SEQ,EQUALS, OPTIONALSEQ, and EQUALSOPTIONAL)used to combine graph patterns, as well as, with a setof functions (e.g. getDURATION, getSTARTTIME)for time-based filtering expressions. The queries arecompiled into event-driven backward-chaining rulesthat can be mixed with other background knowledge.

5.3.8. AnalysisEvents capture the dynamic aspects of a domain

and their efficient representation, processing and anal-ysis are considered key requirements in pervasive andsensor-driven environments. The motivation behind

11http://http://prova.ws/


Table 4Comparison of the event ontologies presented in section 5.3

Ontologies Time

Location

Partici

pation

Mereology

Causal

ity

Correlat

ion

Interpret

ation

Axiomatisat

ion

Modularity

SOUPA√ √ √

– – – – moderate√

Event Ontology√ √ √

limited limited – – limited –Ontonym

√ √ √– – – – moderate

√

SEM√ √ √

limited – –√

limited –Event-Model-F

√ √ √ √ √ √ √rich

√

LODE√ √ √

– – – – limited –

the development and use of ontology-based event mod-els is to provide formal and explicit vocabularies ableto semantically represent and correlate common as-pects of events (e.g., places, people, objects) amenableto reasoning, and thus to high-level interpretation.

This section briefly presented existing event ontolo-gies, focusing on the ontology constructs and designpatterns they provide. The representation of commonaspects of events, such as time, location and participa-tion, is supported by all ontologies. However, the rep-resentation of more complex event relationships, suchas mereological or causal relationships, are fully sup-ported only by the Event-Model-F ontology that pro-vides a rich axiomatisation of ontology design patterns(section 5.3.5). SOUPA and Ontonym follow a mod-ular design with a moderate axiomatisation comparedto the Event Ontology, LODE and SEM ontologies.Among them, only SEM is capable of capturing dif-ferent interpretations of the same event. Table 4 sum-marises the strong and weak points of each ontology.

In addition, this section introduced the notion of se-mantic complex event processing (SCEP), an effort to-wards the combination of Semantic Web technologiesand traditional complex event processing frameworks.The motivation behind this combination is to enrich thepattern-based complex event detection capabilities ofCEP with reasoning over ontology-based event mod-els and background knowledge. In this way, complexevents can be detected and interpreted based on theirhierarchical or temporal relationships to other events,as well as, on their correlations with other relevant con-cepts of the domain, e.g., a relationship between a caraccident event and the location of emergency vehicles.

5.4. Context Reasoning

The higher-level integration of raw context dataand comprehension of their meaning are key pre-

requisites towards understanding a user’s state, be-haviour and surroundings. Espousing the ontology-based modelling paradigm, the low-level context in-formation directly acquired from sensors is translatedto respective ontological class and property assertions.Hence, typical ontology reasoning tasks can be usedto check the consistency of the aggregated set of con-textual assertions and, more importantly, to derivemore complex context abstractions (e.g., recognise auser’s activity based on her current location and ob-jects used) that would otherwise remain implicit. Anabstract ontology-based framework for modelling andreasoning about high-level context is depicted in Fig-ure 10.

In addition to adopting plain ontology-based solu-tions though, a number of approaches have exploredreasoning frameworks that combine ontologies andrules [17,15] in order to cope with OWL’s restrictedrelational expressivity (see Section 3.4). Based on thelevel of interaction that different combination frame-works afford, these approaches can be discriminatedinto three categories: maximising the interaction be-tween the ontology- and rule-based inferences; adher-ing to a tight integration; and adopting a much loosernotion of coupling. In the following, representativeframeworks for each of the three paradigms are dis-cussed.

5.4.1. Ontology-based frameworksThe CoBrA and the Semantic Smart Home (SSH)

frameworks described in Section 4 are both examplesof plain ontology-based approaches to reasoning aboutcontext. In particular, in CoBrA, through a set of on-tologies, useful inferences can be drawn regarding aperson’s location and activity context. For instance,the detection of Alice’s mobile phone in a specific lo-cation entails Alice’s presence [34]. Jena is used tomanage and reason over the knowledge base, whilereasoning is invoked through RDQL that periodically


Pervasive Application

Ontology Reasoner

TBox

ABoxLow-level context information from sensors

Class/property assertions

Complex context abstractions

(e.g. user’s activity)

Pervasive Application

Ontology Reasoner

TBox

ABoxLow-level context

information from sensors

Class/property assertions

Complex context abstractions

(e.g. user’s activity)

Fig. 10. An abstract ontology-based framework for high-level context reasoning from low-level sensory data

queries the knowledge base for the presence of cer-tain context data; e.g., whether a device has been de-tected in a specific room, or who is the owner ofthe device. Once queries return matched results, newassertions about the local context are added to theknowledge base. In an alternative implementation, anobject-oriented rule-based reasoning engine has beendeveloped using Flora-2 [32]. CoBrA also includes aProlog-based meta-rule reasoning component to en-force privacy policies tailored to user needs and pref-erences based on respective sets of user-defined rules.

The ontologies defined within the SSH frameworkcapture knowledge related to physical equipment (suchas sensors and electrical appliances), actions and ac-tivities of daily living (such as watching television andpreparing a meal), location spaces (such as kitchenand living room), people and their roles, medical in-formation, software components as well as tempo-ral information [40]. Modelling activities of daily liv-ing (ADLs) through property restrictions in relation toequipment, location and other types of constraints, al-lows the inference of ADLs on the basis of the as-sertional knowledge made available through sensors.For example, an ADL a is inferred to be an instanceof KitchenADL ≡ ∃hasLocation.Kitchen,if there is sensor evidence that asserts hasLoca-tion(a, kitchen). Further inferences, of higher-level of detail can be drawn as additional sensor de-scriptions are made available. Continuing the exam-ple, the axioms MakeDrink v KitchenADL u∃hasContainer.Container and Cup v Con-tainer, enable to further classify a as an instanceof the MakeDrink class, once the contact sensor at-tached to a cup is activated.

In [201], an ontology-based framework for context-aware applications in the smart home domain is pre-sented using as case study a door lock scenario. Theconcrete situations (i.e. the contextual information di-rectly acquired by sensors) correspond to OWL in-

dividuals and realisation is used to determine intowhich context concepts a specific situation individ-ual falls (e.g., AuthorisedPersonRinging). Evaluationswith three DL reasoners (Racer, RacerPro and Pel-let) demonstrate the feasibility of the proposed frame-work; yet, the examined scenario is rather simple tosafely extend conclusions regarding its evaluation inmuch broader context-aware applications. Towards amore effective engineering the domain knowledge, ageneric methodology for situation modelling has beenproposed in a subsequent work [176]. The methodol-ogy is based on the systematic decomposition of situ-ations according to so called of aspects of interest thatconsider spatial, temporal and acting person criteria,and aims to assist system developers in effectively cap-turing situations at different levels of granularity. Suchmodelling allows to effectively avail of the subsump-tion semantics and derive inferences (though more atcoarser levels of abstraction) when not all pieces of therelevant context information are available.

A more recent example is the OWL 2 based con-text modelling and reasoning architecture presented in[160]. As in the aforementioned initiatives, ontologiesare used not only to represent activities but also rel-evant knowledge that can drive their recognition, in-cluding locations, objects, and so forth. Unlike the pre-vious approaches though, sensors are directly mappedto ontological classes and properties and the sensordata are directly added to the assertional part of theknowledge base. Sensor data are first fed to COSAR[159], where primitive activities through a combina-tion of statistical and ontological reasoning are de-rived. These simple activities, in combination with fur-ther contextual data, once aggregated and processed soas to resolve possible conflicts [1], formulate subse-quently the assertional knowledge, over which OWL2 DL reasoning can be applied in order to recognisemore complex activities. A worth noting feature of theproposed framework is that by exploiting OWL 2’s


support for composition of properties and for qualifiedcardinality restrictions, the captured knowledge is con-siderably more expressive compared to that affordedby earlier frameworks that use OWL 1 DL. Despite therestrictions imposed on the use of the property compo-sition constructor that conditions decidability, the au-thors argue that the use of OWL 2 DL is a satisfy-ing compromise for effectively reasoning, while avoid-ing the technical and semantic complexities confrontedwhen combining ontologies and rules.

5.4.2. Tightly-coupled frameworksAn early example of a context reasoning framework

that combines ontologies and rules can be found inthe Gaia infrastructure. Context information in Gaia isrepresented as first-order predicates, with the name ofa predicate indicating the type of context described.Higher-level contexts can be deduced based on a setof predefined rules that are reevaluated whenever achange occurs [154]. For reasoning in first-order logic,the XSB [166] reasoning engine is used. Quantificationof variables is performed over the specific domain ofvalues, which is finite, ensuring in this way that evalu-ations of expressions will always terminate.

More contemporary examples include the context-aware systems [222,224] that investigate the use ofSWRL rules. In particular, in [222], OWL and SWRLare used to capture and reason over contextual infor-mation about museum visitors, including their prefer-ences and surroundings, in order to provide context-aware recommendation services. Racer and Jess com-prise the system’s reasoning module, allowing for con-sistency checking and taxonomic classification (sub-sumption checking), and the processing of SWRLrules and queries respectively.

Zhang et al. explore the use of SWRL rules in a Se-mantic Web-enabled framework for self-managementin pervasive computing [224]. More specifically, a setof Self-Management Pervasive Service (SeMaPS) on-tologies is used to capture salient notions about per-sons (including their habits and preferences), loca-tions, software agents, devices, malfunctions and re-covery solutions, and quality of service parameters anddynamic context aspects among others; SWRL rulesare used in parallel to capture those parts of com-plex contextual knowledge that cannot be expressedin OWL. Through the Protege-OWL/SWRL APIs, theRacerPro and Jess reasoning engines are used to de-rive context abstractions of higher-level. The infer-ences drawn are subsequently fed to software agentsthat implement BDI reasoning, i.e. reasoning based on

the Beliefs, Desires and Intentions (BDIs) agent model[158]. Provided with enriched contextual knowledgethat would otherwise remain implicit, BDI agents canmake improved decisions and provide better servicessuch as negotiation and consultation.

Despite the use of SWRL rules, reasoning in bothframeworks remains decidable as the existing rea-soner implementations employ inherently the DL-safety notions, supporting in practice a form of re-stricted SWRL.

5.4.3. Loosely-coupled frameworksUnlike the previously described frameworks, loosely-

coupled frameworks keep minimal the interaction be-tween ontology- and rule-based reasoning. Rules areapplied to the consequences derived by means of on-tology reasoning, and affect little the knowledge base.In SOCAM for example, ontology-based reasoning isused to deduce additional knowledge from the contextdata that are directly acquired through sensors; e.g.,based on the transitive semantics of the locatedInrelation, the reasoner can infer that a person is locatedinside the house, provided that she is located in thebedroom contained in it. Once implicit knowledge ismade explicit, first order logic rules are invoked to de-rive higher-level contexts such as sleeping, cookingand watching television [211]. Both reasoning mod-ules are implemented using Jena. A similar rationalehas been adopted in the Semantic Space framework[210]. RDQL queries over the context knowledge baseallow to examine desired contexts so that relevant setsof first order rules can be invoked to derive higher-level contexts. The system uses Jena to implement theforward chaining rules and as the inferred contexts arenot stored into the knowledge base, it avoids the needto handle possible conflicting conclusions.

Ontologies and rules have been also coupled in thehome-based health monitoring and alarm managementsystem proposed by Paganelli and Giuli [140]. Thecontext model consists of four ontologies that cap-ture knowledge about patients (e.g., heart rate andbody temperature), home environmental parameters(e.g. humidity), alarm management (including policiesand contact person information), and social context(e.g., relatives) respectively. Context reasoning is trig-gered whenever a change in the context knowledgebased occurs. Ontology-based reasoning is employedto determine the complex context class to which a spe-cific instance belongs (realisation) and to check theconsistency of the knowledge base once new conclu-sions are added. User-defined rules are employed to


trigger alarms and select notification policies basedon the available biomedical and environmental con-text data. For example, a rule may assert an alarm ac-tivation when the heart rate frequency is less than 40beats/minute and systolic blood pressure is higher than160mm/Hg. Similarly, policy rules are used to assertpolicy-related facts upon detection of alarm activationfacts. Jena has been used to manage and reason overthe application ontologies and rules.

The context-aware access control framework pro-posed in [197] constitutes another example. The frame-work follows a hybrid architecture, combining a DLreasoner (Pellet) and a production rule engine (Jess)in order to apply more expressive context reasoning,such as property path relationships. For instance, ina meeting situation there might be the need to modelthat if the owner of a requested resource is located ina certain place and the resource requestor is located inthe same place, then the two persons are co-located.Such relationships are expressed in terms of produc-tion (if-then) rules whose head and body match classesand properties of the ontology. The output of the rulesis fed into the DL knowledge base to determine thevalue of each attribute given the current context. In thisway, the ontology predicates can be used both in thebody and in the head of rules and thus, the ontologyknowledge may be modified in iterative ontology andrule-based reasoning steps.

Though seemingly effective in practice, it is inap-propriate to compute the consequences of the ontologycomponent first and then apply the rules to the conse-quences [131]. For example, let us assume the follow-ing rules:

InsideKitchen(x) ← Cooking(x)InsideKitchen(x) ← HavingDinner(x)

Given the assertion (Cooking t HavingDinner) :Alice, we know that Alice is either cooking or hav-ing dinner, but we do not know which one is true.However, though either way one of the rules derivesthat Alice is in the kitchen, we cannot deduce it byapplying the rules to the consequences of the ontol-ogy reasoning, as neither |= Cooking(Alice) nor |=HavingDinner(Alice) holds. Detecting and prevent-ing such unwanted behaviour may be straightforwardfor single cases as in the aforementioned example, butbecomes impractical when dealing with large and com-plex knowledge bases confronted in real-world appli-cations.

Susceptibility to incorrect inferences is further ag-gravated by the close-world semantics usually em-

ployed for rules. Such semantics allow the rules to in-trospect the knowledge base and derive conclusionsbased on the absence of information. This induces anon-monotonic behaviour where new inferences mayinvalidate previously derived conclusions. A represen-tative example is given in [160], where a three-roomsmart home, equipped with four sensors, one in eachroom monitoring the presence of people, and one inthe front door monitoring the entrance of people in thehouse, is considered. The knowledge base includes thefollowing definitions:

(1) Room u ¬∃hasOccupant v EmptyRoom(2) EmptyRoom ≡ Room u ¬OccupiedRoom(3) Room(x) ∧ EmptyHome(y) ∧

isInside(x,y) → EmptyRoom(x)(4) ¬EmptyRoom(x) → OccupiedRoom(x)

In the example, the front door sensor asserts the en-trance of one person, yet none of room sensors suc-ceeds to communicate subsequently that a person ispresent. Due to the open world semantics of OWL, rule(4) evaluates to true for all rooms as it cannot be provedthat they are empty. As a result the system ends up in-ferring that there is at least one person present in eachroom.

5.4.4. AnalysisThe aforementioned induce two critical observa-

tions:

1. The combination of ontologies and rules is a keyprerequisite for effectively meeting the expres-sivity requirements when modelling and reason-ing about context in the pervasive domain; how-ever hybrid reasoning schemes that either onlyallow for a poor interaction between the twocomponents or that fail to take into account theparticularities of the co-existence of closed andopen world semantics, may easily lead to incor-rect inferences and an overall undesirable be-haviour.

2. The combination of open- and closed-world rea-soning is desirable when reasoning about con-text in the pervasive domain. The open-world se-mantics are closely related with the capabilityof reasoning over incomplete knowledge, whileclosed-world semantics is needed in order toreason over common conjectures about nega-tive knowledge without having to explicitly statesuch knowledge.


It is worth noting that ontology-based pervasive ap-plications are not the only research domain where theneed to reason, keeping certain parts of the world openwhile closing others, emerges. Similar challenges areconfronted in the semantic understanding of visualcontent [54], and in general in applications that requireintensional reasoning (e.g., natural language process-ing). Furthermore, such seamless integration of openand closed world semantics has been recognised as ahighly challenging yet desirable capability in the Se-mantic Web too, resulting in a number of promisingproposals as discussed in Section 3.4. Their practicalimpact within the pervasive domain remains subject tofuture investigation.

Parallel to this line of research, hybrid architecturesthat orchestrate formalisms for distinct types of rea-soning have emerged. Such investigations begin fromthe premise that different aspects of knowledge mod-elling (temporal, spatial, etc.) impose different require-ments that a single, unified formalism is unlikely tomeet. An example is the Situation Awareness by In-ference and Logic (SAIL) architecture proposed in [8],where a theorem prover and a DL reasoner have beenused to reason over streams of time stamped radar dataand witness reports in order to derive an understandingof the overall situation (e.g. a potentially hostile air-craft approach), and use linear temporal logic to gen-erate appropriate alerts.

5.5. Uncertainty

Pervasive computing offers different challenges forontology developers than traditional applications. Datain pervasive computing environments may be gener-ated by untrustworthy or inaccurate sources and soshould be taken ‘with a grain of salt’. Because compo-nents of a pervasive computing environment deal withthe real world, they come with certain caveats: sensorsin the field are inherently inaccurate, since they couldbreak down; or they could report inaccurately becausethey come up against an unusual phenomenon; i.e. onefor which they have not been designed.

Since these issues must be taken into account whendealing with pervasive systems, it should be possibleto describe the concepts of accuracy, uncertainty, andprovenance with respect to context data and representthem as part of its ontological description. With thesedescriptions in place, particular reasoning mechanismson ontologies need to be designed to support efficientand precise reasoning on the data.

Among the ontology-based models in Section 4,Gaia and CONON undertake this challenge. Gaia triesto capture and make sense of the imprecise and con-flicting data uncertainty inherent in dealing with real-world data [154]. An uncertainty model is developedbased on a predicate representation of contexts and as-sociated confidence values. The predicates’ structureand semantics are specified in ontologies that bene-fit in checking the predicates’ validity; simplifying thedefinition of context predicates in rules; facilitating in-teroperation between different systems; and further re-ducing the possibility of uncertainty when interpret-ing context information. To reason about uncertainty,Gaia employs mechanisms such as probabilistic logic,fuzzy logic, and Bayesian networks, each of which isadvantageous under different circumstances. For in-stance, Gaia uses Bayesian networks to identify ca-sual dependencies (represented as edges) between dif-ferent events (represented as nodes). The networks aretrained with real data so as to get more accurate prob-ability distributions for their event nodes.

CONON associates metadata to each piece of con-text data, describing the quality of each datum. [80]have defined four types of quality parameters: accu-racy, which reflects the estimated error of a measure-ment; resolution, which reflects the smallest perceiv-able element; certainty, which reflects the probabilityof the reading being accurate; and freshness, which re-flects the time a measurement was generated, and itsexpected lifetime. CONON also introduces a depen-dency tag that can be associated with any measure-ment. This allows relationships to be built between dif-ferent contextual information and helps to set down theuser-defined reasoning rules.

Both the above approaches use ontologies syntac-tically as a vocabulary to exchange knowledge basespecified in a probabilistic model. Responding to theneed of modelling imperfect knowledge in the Seman-tic Web, much research has been devoted to extend-ing formalism and reasoning services so as to han-dle uncertain and/or vague information. Representativeexamples including among others fuzzy extensions ofDLs [182,181], OWL [180,20] and SWRL [214], andprobabilistic extensions such as PR-OWL [52,30] andBayesOWL [58]; for an extensive overview the readeris referred to [183]. Further relevant proposals includethe pattern-based approach for representing and rea-soning with fuzzy knowledge [202], and the generic,formalised approach for managing uncertainty pro-posed by Dividino et al. [59]. Few works, however,have explored the applicability of such initiatives in the


domain of pervasive applications; an example is theapproach presented in [45], where fuzzy reasoning isused to provide personalised mobile services based onsituation awareness.

Missing data is another source of uncertainty whenreasoning about context: a miss (or inaccurate) detec-tion of low-level context information may easily leadto irrecoverable failures in the inference of higher-levelcontext abstractions. One possible solution is to modelthe interpretation of perceptual data as inference to thebest explanation using abductive reasoning [172,146].Romero et al. [74,75] investigate this idea in the con-text of an ontology-based surveillance application. Aset of ontologies are used to capture context at increas-ing levels of abstractions, including tracking knowl-edge, scene objects and activities. Once the low-levelcontext acquired from visual sensors is translated intoABox assertions, abductive rules are applied to derivemissing facts and trigger the derivation of higher-levelcontext descriptions. No information is provided what-soever about the computational framework used to im-plement the abductive reasoning and the preferencecriteria used for selecting explanations. Acknowledgedas a mode of reasoning that is inherent in a plethora oftasks, much research has been devoted to understand-ing abduction. For a detailed account on the potentialof abductive reasoning in DLs, the interested readermay refer to [63,105].

A formal model based on defeasible logic is pro-posed by Bikakis et al. to support reasoning with im-perfect context in ambient computing environments[16]. Extending the Multi-Context Systems modelwith non-monotonic features, the proposed frameworksupports reasoning in cases of missing context knowl-edge. Potential inconsistencies are resolved by meansof an argumentation framework that exploits contextand preference information that expresses confidenceon the contexts considered. The propositional repre-sentation of context knowledge may not allow a di-rect integration with ontology-based context reasoningframeworks; yet possibilities for interesting hybrid ar-chitectures emerge where contextual assertions can beselectively translated into equivalent grounded formu-las.

5.5.1. AnalysisAt present most of ontology-based models in perva-

sive computing community are still at the stage of us-ing semantic annotations to tag different quality mea-sures. One of the future directions in dealing with un-certainty is to use the ontologies that are tightly inte-

grated with probabilistic or fuzzy reasoning. Abduc-tive reasoning is also worth of further investigation inthat it can help to detect errors (e.g., missing or inac-curate data) which can be used as a feedback to re-tunethe system.

5.6. Semantic Service Discovery

One of the core problems in pervasive computing isthe question of how to support the introduction of newdevices into the environment. As these devices maybe unaware of the configuration of the environment orthe services available they must undergo a discoveryor matchmaking process to best integrate themselves.According to [110], discovery is a process of retrievingservices that are able to fulfil a task, and matchmak-ing is a process of automatically matching semanti-cally annotated services with a semantically annotatedrequest.

Gaia uses DAML+OIL to achieve semantic discov-ery as it supports some of the operations required forsemantic discovery like classification and subsump-tion. It also allows the definition of relations betweenconcepts. The use of ontologies and semantic discov-ery replaces scripts and ad hoc configuration files thatwere used in Gaia previously. Each entity is associatedwith a DAML+OIL description that describes its prop-erties. The ontology server poses logical queries in-volving subsumption and classification of concepts tofind appropriate matches. Other entities in the environ-ment may query the ontology server to discover classesof components that meet their requirements. Match-making uses ontologies to determine a set of conceptsthat fulfil the intersection of the requirements of two ormore parties, such as a supplier and a consumer usingthe matching algorithms described in [199]. The resultis a set of classes that are semantically compatible tothe query class.

Christopoulou et al. [43] uses their GAS ontologyfor discovery in the event of a component failure. Allcomponents have ontologies describing their interfacesand available services. If a connection (or synapse) be-tween two or more components is broken, the ontologymanager attempts to find an alternative component thatoffers the same service (i.e., that has the same ontolog-ical description) to repair the connection.

The DAML-based Web Service ontology (DAML-S) supplies web service providers with a core set ofmarkup language constructs for describing the prop-erties and capabilities of Web Services [28,29]. It isused in combination with additional ontologies, which


Service Profile

Profile

&process;#Parameter

&expr;#Condition

&process;#Result

&process;#Input

&process;#Output

hasParameter

hasPrecondition

hasResult

hasI

nput

hasO

uput

ServiceCategory

ServiceParameter

&xsd;#anyURL

&xsd;#anyURL

Process

Thing

serviceParameter

serviceCategory<r

dfs:

subP

rope

rtyO

f"#

hasP

aram

eter

"/>

serviceClassification

serviceProduct

serviceName

textDescriptio

n

has_process

serviceParameterName

sParameter

categoryName

taxonomy

code

value

DatatypeProperty

SubClass/Property

ObjectPropertyLegend

Fig. 11. Selected classes and properties of the profile in OWL-S [115]

organise the concepts appearing in the DAML-S de-scriptions. The use of DAML-S renders the semanticsof the descriptions machine comprehensible; thereforeit enables intelligent agents to discover, invoke andcompose web services automatically. CoOL extendsDAML-S with the service context to offer a more for-mal description of a service’s contextual interoperabil-ity. The service context consists of two parts: the con-text obligation, which specifies the obligations of a ser-vice in terms of the context of its execution; and thecontext binding, which is used to establish a virtuallink from an atomic process of a service to a specificaspect of the context [187]. This formal semantic ser-vice description offers a common understanding of therelations between services and their associated con-texts [184]. This facilitates context-awareness and con-textual interoperability during service discovery andexecution.

[13] introduces a dedicated Service Discovery Pro-tocol (SDP) that enables advertising and discoveringservices in pervasive environments according to thesemantics of networked services and of sought func-tionalities. The approach is based on the Amigo-Slanguage that extends OWL-S [115] (the successorof DAML-S based on OWL) with new classes andproperties. Amigo-S enriches the profile-based anno-

tation paradigm of OWL-S for describing inputs, out-puts, preconditions and effects (IOPEs) by incorporat-ing also a number of capabilities for each service. Thecapabilities are considered either as provided, i.e. ca-pabilities that are supported by the service, or required,i.e. capabilities that are needed by the service. [112]defines five possible ‘degrees of match’: exact, plug-in,subsume, intersection and disjoint. These groups indi-cate ‘how good’ a match is.

The OWL-S based approaches cannot support finer-grained service matching where priorities or weightson individual requirements of a service request needto be taken into account. Also they lack an appro-priate criterion to approximate and rank the avail-able service with respect to a given request [11].To address these issues, [11] extends the request de-scription with a priority value to indicate the rele-vant importance of the individual requirements in arequest; e.g., Req v (= 1hasDescription.RD) u(= 1hasPriority.PriorityV alue). The matchmak-ing process starts with checking if all mandatory re-quirements are satisfied by a service. If all of them aremet, then the service will be evaluated through approx-imate matching, where the similarity between this ser-vice and the request will be determined depending onthe semantic deviation of the expected value in request


and the available values in the service description forthe same requirement. A similarity score will be as-signed, which will be used to rank the candidate ser-vices later.

5.6.1. AnalysisSemantic service discovery is a critical function in

pervasive computing that enables ad-hoc associationsbetween service providers and consumers to be estab-lished [126]. Gaia, CoOL, and SDP have demonstratedthat semantic service discovery is better than tradi-tional, rigid, syntactic service matching. OWL-S, anexcerpt of whose classes is shown in Figure 11, is cur-rently the most popular of the available technologiesfor semantic service description. As aforementioned,though it suffers certain limitations when it comes tofine-grained service mathing, where priorities on indi-vidual service attributes need to be addressed. Otherrepresentative semantic service description languagesinclude WSML, WSDL-S/SAWSDL, Monolithic DL-based, and DSD for non-logic-based service IOPEprofile [106]. Each of these languages has its ownstrengths and particular application areas, and the cur-rent challenge is how to enhance the expressiveness ofservice annotations so as to increase the capabilitiesfor searching for candidate services.

Approximate matching is another interesting topic,as dealing with vague or incomplete descriptions aboutthe functionality of services or user preferences is notuncommon. Such considerations are not exclusive tothe pervasive domain, and have attracted significantbody of research in the semantic matchmaking com-munity [106,70].

5.7. Privacy and Trust

Privacy and trust are of paramount concern, if per-vasive computing applications are to become popu-lar outside the laboratory; yet, only a limited numberof the reviewed pervasive frameworks have explicitlyconsidered these aspects.

One such example is the policy-based approach im-plemented in CoBrA, where the SOUPA policy ontol-ogy [36] is used to enforce user privacy. Using thisontology, users can define customised policy rules topermit or forbid different computing entities to accesstheir private information. The policy reasoning algo-rithm uses a DL inference engine and the DL con-structs of OWL to decide whether an action for access-ing some user private information is permitted. Since itis often infeasible to define explicit policy rules for ev-

ery individual action in a domain application, CoBrAuses meta-policies that determine behaviour when pol-icy rules are not defined. These meta-policies can beeither conservative, in which case the system assumesthat all actions are forbidden, or liberal, in which caseit assumes all actions are permitted. A CoBra proto-type, implemented by Chen et al. [33] for an intelli-gent meeting room application, was used to demon-strate that the SOUPA policy ontology, and its asso-ciated algorithms enable to develop intelligent agentsthat can provide user privacy protection in a pervasivecontext-aware environment.

A policy ontology is also proposed by Paganelliand Giuli within their context-aware framework forat home monitoring of patients with chronic diseases[140]. This ontology however addresses the manage-ment of alarm notifications, specifying among otherspriority information about which care person shouldbe contacted first, and does not cover privacy and trustaspects.

5.7.1. AnalysisAs we wish to communicate increasing amounts of

(personal) information between services, the topics ofsecurity, privacy, and access control become more andmore critical. To harvest the advantages that SemanticWeb technologies bring in communicating and sharingknowledge across systems, a consensual policy ontol-ogy is highly desirable in order to capture in a stan-dardised manner aspects pertinent to privacy and ac-cessibility, such as who is granted access to whichparts of data and for what purpose (e.g., reading orediting). The need towards such standardised methodsfor expressing and exchanging privacy policies, hasbeen widely acknowledge within the Semantic Webcommunity itself (e.g. ) and is closely related with thequest and recent efforts for formalising the semanticsof provenance information (see Section 6.3).

5.8. Summary

This section explicated the use of Semantic Webtechnologies in the domain of pervasive, sensor-drivenapplications. The discussion has been structured alongthe five key requirements, as identified in section 2,namely: i) conceptual modelling of context (rangingfrom raw sensor data to higher-level context and eventabstractions), ii) reasoning about context, iii) uncer-tainty management, iv) service discovery and v) pri-vacy and trust. Besides outlining strengths and weak-nesses of the proposed approaches, relevant results


Table 5Overview of Semantic Web technologies use in pervasive applications with respect to key requirements.

Requirement SW-empowered approach Added value Further research inquiries

Modelling andrepresentingcontext

ontologies for describing sensordata, who/where/what/when infor-mation, events; rules for higher re-lational expressivity

facilitate knowledge sharing, reuseand exchange; rich expressivenessfor capturing complex semantic re-lations

upper ontologies; comparative as-sessment of existing ontologies

Reasoning aboutcontext

DLs reasoning services; rule-basedreasoning; reasoning with ontolo-gies and rules

inferring implicit knowledge, con-sistency checking, reasoning aboutincomplete knowledge

scalability; combining open- andclosed-world reasoning; reasoningunder inconsistency; temporal/real-time reasoning; hybrid reasoningframeworks

Uncertainty han-dling

extensions for fuzzy/probabilisticsemantics (e.g., Fuzzy DLs, PR-OWL); non-standard inferencing(e.g. abduction in DLs ABoxes)

reasoning over imprecise/vagueknowledge; reasoning to hypothe-ses

scalability; seamless integration ofuncertainty; unified handling of dif-ferent types of uncertainty

Service discovery standardised (DAML-S, OWL-S)and custom service-oriented on-tologies

semantic discovery and matchmak-ing of services

scalability; adaptive semantic ser-vice composition and execution(e.g. due to service unavailability)

Privacy and trust custom policy ontologies facilitate sharing and exchanging ofprivacy policies

standardised policy/provenance on-tologies

within the Semantic Web community that have notbeen yet explored have been discussed, sketching pos-sible directions for further investigations.

Table 5 summarises the main observations. The useof Semantic Web technologies induces a number ofstraightforward benefits, as a direct result of the ad-vocated explicit semantics and well-defined reason-ing services. Issues open to further investigation areto a large extent inherited by challenges that remainopen in Semantic Web research too. Efficiency of rea-soning is crucial, as the need to draw complex in-ferences from millions or billions of pieces of infor-mation in real-time, is not restricted in the domainof pervasive systems alone. The seamless integrationof open- and closed-world reasoning is another suchhighly desirable feature and subject of active research,and the same holds for investigations into the practicalmanagement of imperfect (uncertain, vague, missing,noisy) knowledge. Last but not least, throughout thestudy of SW-based pervasive frameworks it was evi-dent that relevant insights often permeated only par-tially, and in a fragmented manner, the borders be-tween the two research communities.

6. Challenging Issues

In the above sections, we have discussed key re-quirements in pervasive computing and have analysedthe benefits and potentials that are induced by usingSemantic Web technologies, along with future research

inquiries and directions. This section analyses open re-search issues relevant to temporal features, dynamism,provenance, and programming that need to be also in-vestigated towards the seamless integration of Seman-tic Web technologies and pervasive computing.

6.1. Temporal Features

The role of temporal information in the modellingof sensed data is often simplified in the design of datamodels, systems software, and application APIs. Typi-cally, sensor values are recorded in conjunction with atimestamp. However, despite its universality as a mod-elling concept, the use of timestamps in isolation im-plicitly restricts a data model to representing only cur-rent state; that is, a timestamp has an implicit dualmeaning: it is both the time at which a statement wasasserted in the model and the time at which the state-ment should be interpreted as being true. This distinc-tion is often unimportant, however the ability to sep-arately annotate data with temporal extent, i.e., a timeinterval, caters explicitly for the latter case.

Further to this, the use of both an update timestampand temporal extent in concert provides a useful facil-ity for modelling both historical and predictive state.This is often useful, for example: summarising a highvolume of sensor readings over a period of time ispreferable to removing it entirely in the case wheredata will later be analysed offline, or representing aperson’s predicted future locations based on calendardata. Modelling time as an orthogonal concern allows


such cases to be handled without requiring special in-dicators in a data model to indicate that informationhas been treated thusly.

Despite strong vocabulary support for represent-ing temporal features, as discussed in section 5.2, thetriple-based nature of the RDF model is ill-suited tothe representation of data’s temporal properties, wherethe requirement is to annotate sets of triples. At leastthree possible strategies for overcoming this limitationare available to data modellers: using RDF’s reificationvocabulary, externally generating identifiers for state-ments, or use a non-standard extension to the RDFmodel.

To utilise RDF’s reification vocabulary, each tripleis expanded as a ‘reification quad’12, with the resul-tant statement identifier associated with temporal in-formation. Generating a statement identifier externallyfollows similarly, but avoids the introduction of redun-dant information into the model at the expense of re-quiring non-standard tooling to generate and resolvesuch identifiers.

Temporal extensions to RDF, all notionally based onexpanding the triple model to a quad model, have beenexplored in the literature. Lilis et al. [113] introduceMultidimensional RDF, an extension to RDF designedto express the temporal semantics of a collection ofcultural artefacts. Time is used as a contextual specifierto control whether or not an RDF triple should be con-sidered to be present in a graph at a given time point orinterval. Gutierrez et al. [81] provide the semantics fortemporal RDF graphs, introducing the notion of a tem-poral triple, an RDF triple with a temporal label, anda temporal graph made from a set of temporal triples.The authors present the concepts of graph slices andgraph snapshots, which allow for the description of thecollection of triples that hold during or at a given in-terval or instant. This foundational work in modellingdynamic data using RDF, influences the design of thetemporal component of our sensor data model.

Based on the notion of temporally-extended RDF,Pugliese et al. [153] introduce the tGRIN index struc-ture that builds a specialised index for storing temporalRDF in a relational database based on temporal as wellas the structural closeness of triples, while Tappolet etal. [191] present a method for building a meta-index

12The term ‘reification quad’ refers to a set of 4 triples that asso-ciate an identifier with the type rdf:Statement, and three prop-erties, rdf:subject, rdf:predicate, and rdf:objectwith values corresponding to the original triple.

for the validity of named graphs based on Elmasari etal.’s Time Index [67].

To facilitate querying of temporal RDF, Tappo-let et al. introduce a syntactic sugar extension toSPARQL to facilitate the querying of graphs valid ata certain time point [191]. McBride et al. [118] at-tach temporal extent to objects, upon which they pro-vide a syntactic sugar in query language that mapsto SPARQL. Rodriguez et al. introduce a schemefor storing, indexing and querying temporal data de-signed for sensor networks [2]. Perry et al. [148] in-troduce the syntax and semantics of an extension toSPARQL, called SPARQL-ST, designed for the exe-cution of spatiotemporal queries. O’Connor et al. de-velop a lightweight temporal model to encode databased on the valid-time dimension, upon which theydevelop extensions to their SWRL-based OWL querylanguage SQWRL [137,138]. Their query library sup-ports Allen’s relations and provides functions forgrouping query results such that the filters first, first-n,last, last-n, and nth can be applied within a query.

In addition to solutions that are solely based on tem-poral extensions of Semantic Web technologies, hybridframeworks that combine ontologies with temporally-aware formalisms have also been investigated. In theambient computing framework proposed by Patkos etal. [145] for example, the inherent temporal reasoningcapabilities of the Event Calculus [107] are utilised.Rule-based reasoning is used, on top of context mod-elling ontologies, to infer complex contexts from rawcontext data. In parallel, causality reasoning is em-ployed to capture and reason over preconditions andeffects of actions and events, based on the Event Cal-culus theory. A main advantage, in comparison to plainrule-based approaches, is the inherent notion of timein the Event Calculus that allows to establish a lineartime ordering and hence infer in which intervals cer-tain conclusions hold, while re-evaluating event pat-terns as time progresses. The situation awareness ar-chitecture presented in [8], is another example wheredifferent formalisms and available reasoners have beencombined to enable efficient inference over data thatchange over time. Sensor observations are first ag-gregated by means of if-then rules, and subsequentlyfed to the semantic interpretation layer, where a DLreasoner is used for situation assessment. The SCEPframeworks described in section 5.3.7, coupling the in-herent real-time reasoning capabilities of CEP engineswith rich ontological semantics, are yet another exam-ple of investigations towards the efficient handling oftemporal semantics.


The aforementioned indicate a significant gap be-tween the currently available Semantic Web technolo-gies and the need for native temporal data modellingand reasoning. Relevant suggestions exist in researchprototype form, but considerable efforts are requiredbefore achieving standardised solutions that will inturn ensure necessary tool support for efficient reason-ing over temporal data and management of temporalqueries. Within this quest, hybrid frameworks presentappealing features, as through the combination of dif-ferent formalisms and reasoners, extended representa-tional and reasoning capabilities can be achieved; theseamless and scalable integration of the heterogeneousmodules inevitably poses its own challenges.

6.2. Dynamism

Distinct from the need to model the temporal fea-tures of dynamic data are the challenges related stor-ing, reasoning over, and accessing large volumes ofhighly-dynamic, distributed information.

Whether a centralised or decentralised setup isadopted, as the volume of sensor data increases, it be-comes infeasible to store a permanent record of gener-ated states. A typical solution might be to discard theoldest data from memory when the maximum capacityis approached. However, it does not necessarily followthat the oldest data stored is the least useful. One pos-sible solution might be to require that the data modelcaptures properties describing data dynamics such thatpolicies may be defined to prioritise the discard of datathat is asserted frequently but has a slow rate of change(e.g., ambient temperature) over data that is frequentlyasserted but changes rapidly (e.g., a user’s coordinatelocation) or pseudo-static data that is infrequently as-serted (e.g., building layouts) [178].

Data dynamics also play a role in determining thetypes of reasoning strategies that should be adopted.There is a need to track derivations such that it may bedetermined when an inference no longer holds due tomodifications to or invalidation of the data upon whichit is predicated. Structuring the data model so that datais temporally qualified (and hence inferences are alsotemporally qualified) forms part of a solution to thisissue, but requires the incorporation of the appropri-ate temporal semantics within existing reasoning tech-nologies.

The process of reasoning over a large data modelcan be a performance bottleneck, however knowledgeabout the dynamics of data can inform an appropriatereasoning strategy. For example, choosing to reason on

static data as it is generated, adding the inferred knowl-edge directly to the model, while performing reason-ing on highly dynamic data only at the point where anapplication query is executed, restricting reasoning tothe smallest amount of volatile data that will produce acorrect answer. Based solely on data’s temporal prop-erties, it may be possible to devise a general schemeto partition, reason on, and integrate data over severalstages so as to optimise reasoning.

6.3. Provenance

Many of the issues above can be subsumed un-der the notion of provenance, i.e. the capture, rep-resentation and manipulation of knowledge relevantto data creation, ownership, transformation and other‘lifecycle’ issues. Provenance appears in many guises.The Dublin Core vocabulary [55] is perhaps the best-known, providing terms for asserting authorship andother property rights over digital objects. More re-cently a task group of the World Wide Web Con-sortium has been standardising the Open ProvenanceModel [127] for asserting more general provenancemetadata.

Sensor-driven systems are more directly affectedby data provenance than programs in many other do-mains. A sensor system must make decisions using in-put data that is known to be inherently noisy, impre-cise, inaccurate, untimely and infrequent, with the de-gree of each source of error perhaps varying signifi-cantly over time. An immediate consequence of thisis that sensor-driven systems cannot be directly con-nected to their input data streams, or respond directlyto them: to do so is to allow the system to respond totransient noise. Another way of putting this is that in-dividual data elements are evidence of fact rather thanbeing facts themselves, and must be fused with otherdata (from the same or different sources) to build aconsensus of the state of the environment being sensed.

While sensor fusion is commonplace in much ofengineering, many of the approaches make strong as-sumptions about the nature of the data streams: for ex-ample they are from homogeneous sensors looking atthe same phenomenon and with well-known samplingfrequencies. Such techniques do not generalise wellto semantically-enriched systems with highly hetero-geneous data sources processed using symbolic rea-soning. Conversely, homogeneous data can be handledeasily within reasoners, allowing semantic technologyto encompass many sensor fusion tasks.


Reasoning over sensor data is affected by a num-ber of provenance factors of data streams and individ-ual data. A trivial example is that the reliability of atemperature estimated from data that is several hoursold may be assumed, all other things being equal, tobe less than that inferred from data collected withinthe previous minute. Sensor types (and even individ-ual sensors) give rise to data with given provenancein terms of the observation frequency, precision andthe like, which can be captured generically or usingspecial-purpose markup languages like SensorML (seesection 5.1). The provenance here is associated withthe sensor but attaches to the data (stream or items)produced by the sensor.

More generally, a system may be interested in theentire lifecycle of a data stream including the transfor-mations that have been applied to it. Suppose again weare receiving a stream of temperature values which weintend to use in some decision process. It may matterwhether those values are ‘raw’ (and subject to raw sen-sor noise), and have been processed to remove obvi-ous outliers, or have been smoothed using an a priorior learned smoothing function. It is therefore vital forsystems where data comes with provenance that thisprovenance is manipulated and maintained along thedata pathway. For many scientific systems the statisti-cal properties of a data stream are as important as thedata itself. Extensive in-system processing of data isnot necessarily a problem, if it is clear to the end-userthat such processing has occurred and to what effect– otherwise any further statistical calculations are ren-dered suspect.

Provenance is increasingly recognised as a crucialelement of the Semantic Web, as it enables inferencesto be drawn conditional to whether information shouldbe trusted and how it should be reused and integratedwith other diverse information sources. The lack, how-ever, of a standard model for capturing, interchangingand reasoning provenance metadata is a significant im-pediment to realising applications where the trustwor-thiness and the quality of the statements is at issue. TheProvenance Working Group [151] is an ongoing efforttowards the standardisation of an interchange core lan-guage (PROV Ontology [149]) for publishing and ac-cessing provenance metadata, drawing on existing vo-cabularies and ontologies [127,150,152]. In parallel,recent research studies have proposed approaches thatenable the handling of provenance while using and rea-soning about information and resources in open andcollaborative environments [22,139,196].

6.4. Programming

Lastly we come to the practicalities of program-ming.

Semantic structures like RDF allow rich encodingsof data. Coupled with OWL, SPARQL, ontologicaland other reasoners, it is possible to answer complexqueries by traversing the knowledge graph. The emer-gence of standard, re-usable reasoners significantlysimplifies the use of semantic technologies within pro-grams.

The integration of these tools remains superficial,however. From a programming perspective, a knowl-edge graph is simply a collection of edges labelledwith strings and URIs, possibly with some additionaltyping at the endpoints supplied by XML Schemata,and with some structure on the relationships providedby the accompanying ontologies (if any). The gener-ality of these structures forces programmers to pro-vide any additional machinery by hand, without theassistance of a compiler, type system of other toolingwith which we are familiar in the building of complexsoftware systems. In an effort to simplify the integra-tion of Semantic Web technologies into existing soft-ware architectures and languages, a plethora of Ap-plication Programming Interfaces (APIs) for workingwith RDF/OWL ontologies have been developed, suchas Jena API [99], OWLAPI [90], dotNetRDF [62],RDFReactor [207] and Sesame [171].

Use of semantic technology implies considerable fa-miliarity with the tools of XML, notably namespaces,that obscure rather than illuminate the underlying in-formation being encoded. Data so encoded is not heldin this form in memory, and so must be translatedfor storage and exchange: a non-trivial process in thepresence of shared sub-structures and update. State-of-the-art ontology editors, such as Protégé [136], Top-Braid Composer [198], and NeOn Toolkit [134] offercomprehensive support for developing and validatingRDF/OWL ontologies. Moreover, scalable and queryefficient RDF repositories, such as OWLIM [18], Al-legroGraph [3], and OpenLink Virtuoso [68], and aswell as, database-to-RDF mapping tools such as D2RQ[86,66], enable data to be exposed and shared on theWeb according to the principles of Linked Data [19].

Entities represented within a knowledge graph maybe classified by the ontology according to the informa-tion present about them – classifying a person as anemployee given the presence of an employee numberand/or a line manager within the same organisation, forexample. This is superficially the same as a program-


ming language type system, with the difference that anentity’s class may change unpredictably and at unpre-dictable times through the actions of other agents onthe knowledge graph, and it can be hard to predict ex-actly which such changes will have such an effect. Thisdestabilises a program’s view of the knowledge in thegraph.

Use of a reasoner sits outside the programming lan-guage, in the same as does SQL when accessing adatabase: a SPARQL query is a string that returns ar-rays of other strings, and is not type-checked or manip-ulated using dedicated programming constructs. Thiscomplicates the formation and checking of complexqueries, again from a lack of supporting tooling. Someof the aforementioned ontology editors provide only alimited support for advanced query formulation, e.g.,type-checking.

Finally, the Semantic Web may present steep learn-ing and commitment curves, particularly for novicescoming from other research areas. In order to performeven simple tasks, a developer must master a widerange of perhaps unfamiliar technologies. Moreover,an organisation must commit to these technologies andtheir concomitant costs ahead of time. This raises abarrier to deployment by increasing the risk that a sys-tem may not generate the expected benefits while in-curring a substantial up-front cost.

As with any technology, the decision to use Seman-tic Web technologies, does not come without a price,and one needs to ascertain sufficient value on its ad-vantages – open, standards-based representation, easyexchange and integration – to make it worthwhile. Itis undoubtedly attractive to be able to define a struc-ture for knowledge that exactly matches a chosen sub-domain, to describe the richness of this structure, andto have it compose cleanly with other such descrip-tions of complementary sub-domains defined indepen-dently – and to be able to exchange all this knowledgewith anyone on the web. But this flexibility comes witha cost and (often) no obvious immediate, high-valuebenefits.

In many ways, these issues are an inevitable conse-quence of the differences in the application domains ofsemantic technology and programming languages. Theformer addresses the open, extensible, scalable, dis-tributed mark-up of data. The latter addresses almostexactly opposite issues, focusing on close specificationof algorithms and data structures to share common dataand functionality.

Sensor-driven systems add few, if any, specificpoints to this general discussion: the issues apply to all

programming with semantic structures. Sensor-drivensystems add the necessity to embrace the noise and er-rors inherent in data streams, and to capture and makedecisions that propagate this uncertainty throughoutthe system. These are not issues that mainstream pro-gramming languages provide structures for, althoughit is possible to build programs that (for example)maintain the error bars on results derived from pro-cessing inputs that themselves have error bars. Com-bining programming with uncertainty and program-ming with data that adhere to an ontological struc-ture, seems to be two new frontiers for programmingsystems design opened-up by semantically-enrichedsensor-driven systems.

Perhaps the chief impediment to such designs comesfrom the challenged nature of the platforms them-selves. Sensor-driven systems place a significant amountof their functionality on devices with extremely con-strained memory, computation, and communicationcapabilities, which are not straightforwardly subjectto Moore’s-law-driven improvements in performance.A naïve deployment of Semantic Web technologies tosuch platforms is doomed to failure, but there seemsto be no a priori reason why versions optimised forrestricted domains might not be possible, and mightnot interoperate easily with the wider universe of web-enabled components.

7. Discussion and Concluding Remarks

In this paper, we reviewed the landscape of the ap-plications of Semantic Web technologies to the domainof pervasive, adaptive, sensor-driven systems. Many ofthe features underpinning the Semantic Web, and inparticular, the ability to formally capture and reasonover rich, semantic interconnections between data, inan open and extensible way, are well-suited to perva-sive computing. At the same time however, a numberof other important aspects, such as lack of native sup-port for representing and reasoning over temporal orimperfect knowledge, remain under-articulated.

The key benefits of Semantic Web technologiesare rather straightforward in terms of modelling andreasoning about context. As afore-described, ontolo-gies have been proposed for describing the four Ws(when, where, what, who) that characterise contextualknowledge, complex events and sensor data, facili-tating the unambiguous sharing and understanding ofknowledge across heterogeneous and distributed plat-forms, devices and services. What still remains un-


addressed, acknowledging the different world-viewsthat different ontologies inevitably serve, is the defini-tion of commonly-agreed (possibly upper) ontologiesthat would enable the standardised description of perti-nent context aspects; the conceptual overlaps betweenthe proposed ontologies, underline further this need.Moreover, since not all of the provided modelling ca-pabilities have been directly applied in pervasive appli-cations (e.g. modelling of composite events and theirdependencies that many of the event ontologies sup-port), it is time to systematically assess their applica-bility, possibilities for their combined use and/or (par-tial) alignment, as well as aspects that require a moreelaborate coverage.

In parallel, the deployment of Semantic Web tech-nologies has demonstrated the intrinsic relation be-tween automated reasoning and the high-level inter-pretation of context data, where the integration ofstructured domain knowledge is a prerequisite. Besidesuseful insights on the need to combine ontologies withrules, the reviewed literature sketches a number of de-sirable, yet highly challenging questions. Prominentones include the need to provide native support for rep-resenting and reasoning over temporal knowledge, in-corporating provenance into reasoning, and managinguncertainty. An indispensable requirement, underlyingall of aforementioned research directions, is computa-tional efficiency. Ensuring reasoning efficiency is cru-cial, as the need to draw inferences in real-time, frommillions or billions of pieces of information, in anopen pervasive environment, already challenges exist-ing reasoning engines, even though considering onlydeductive inference over crisp, consistent knowledgebases.

Subject to further enquiry is also the overall inte-gration of Semantic Web technologies within program-ming frameworks suitable for software engineering in-the-large. The Semantic Web at present is essentiallya collection of fragments lacking a whole. This is per-haps an inevitable consequence of an architecture de-signed for such a broad spectrum of application do-mains, but it nevertheless increases the risks and costsassociated with applying the technologies to pervasivesystems. This is especially the case when the targetplatforms for much of the functionality are challengedin terms of their memory, computation and communi-cations capabilities. It is important to remember, how-ever, that the Semantic Web is essentially a tool ofmodelling and exchange, and only secondarily a tool ofimplementation: compact representations of informa-tion and reasoning that comply to the underlying meta-

model of the Semantic Web seem eminently possibleand deserve future exploration.

Summing up, the ability to exchange models anddata, to reason openly, to capture an extending set ofdata and metadata, and to interact with other web-enabled elements, all encourage the view that bas-ing future pervasive and sensor-driven systems aroundthese technologies would lead to significant improve-ments in interoperability and semantic clarity – if thedisparate elements can be integrated into a frameworkappropriate for system developers. Such an integra-tion would allow innovation to proceed more rapidlyand soundly, bringing Weiser’s vision of seamless,integrated pervasive and sensor systems significantlycloser.

References

[1] A. Agostini, C. Bettini, and D. Riboni. Hybrid reasoning inthe care middleware for context awareness. Int. J. Web Eng.Technol., 5(1):3–23, May 2009.

[2] Y. L. Alejandro Rodríguez, Robert McGrath and J. Myers.Semantic management of streaming data. In Proceedings ofthe 2nd International Workshop on Semantic Sensor Networks(SSN09), volume 522 of CEUR Workshop Proceedings, pages80–95, 2009.

[3] AllegroGraph RDFStore Web 3.0’s Database. http://www.franz.com/agraph/allegrograph/.

[4] J. F. Allen. Towards a general theory of action and time. Ar-tificial Intelligence, 23(2):123–154, 1984.

[5] D. Anicic, P. Fodor, S. Rudolph, and N. Stojanovic. Ep-sparql: a unified language for event processing and streamreasoning. In Proceedings of the 20th international confer-ence on World wide web, WWW ’11, pages 635–644, Hyder-abad, India, 2011. ACM.

[6] D. Anicic, S. Rudolph, P. Fodor, and N. Stojanovic. Streamreasoning and complex event processing in etalis. SemanticWeb - Interoperability, Usability, Applicability, 1:1–5, 2011.

[7] G. Antoniou and F. v. Harmelen. A Semantic Web Primer, 2ndEdition (Cooperative Information Systems). The MIT Press,2 edition, 2008.

[8] F. Baader, A. Bauer, P. Baumgartner, A. Cregan, A. Gabal-don, K. Ji, K. Lee, D. Rajaratnam, and R. Schwitter. A novelarchitecture for situation awareness systems. In TABLEAUX,pages 77–92, 2009.

[9] F. Baader, D. Calvanese, D. L. McGuinness, D. Nardi, andP. F. Patel-Schneider, editors. The Description Logic Hand-book: Theory, Implementation, and Applications. CambridgeUniversity Press, 2003.

[10] F. Baader and U. Sattler. An overview of tableau algorithmsfor description logics. Studia Logica, 69(1):5–40, 2001.

[11] A. Bandara, T. Payne, D. D. Roure, N. Gibbins, and T. Lewis.Semantic resource matching for pervasive environments: Theapproach and its evaluation. Technical report, Faculty ofPhysical and Applied Science, University of Southampton,2008.


[12] N. Baumgartner and W. Retschitzegger. A survey of upperontologies for situation awareness. In In Proceedings of the4th IASTED International Conference on Knowledge Sharingand Collaborative Engineering, pages 1–9, St. Thomas, USVirgin Islands, November 2006.

[13] S. Ben Mokhtar, A. Kaul, N. Georgantas, and V. Issarny. Ef-ficient semantic service discovery in pervasive computing en-vironments. In Proceedings of the ACM/IFIP/USENIX 2006International Conference on Middleware, Middleware ’06,pages 240–259, Melbourne, Australia, 2006. Springer-VerlagNew York, Inc.

[14] T. Berners-Lee, J. Hendler, and O. Lassila. The semantic web.Scientific American, 284(5):34–43, May 2001.

[15] C. Bettini, O. Brdiczka, K. Henricksen, J. Indulska, D. Nick-las, A. Ranganathan, and D. Riboni. A survey of context mod-elling and reasoning techniques. Pervasive and Mobile Com-puting, 6(2):161–180, 2010.

[16] A. Bikakis and G. Antoniou. Contextual argumentation inambient intelligence. In Proceedings of the 10th InternationalConference on Logic Programming and Nonmonotonic Rea-soning (LPNMR ’09), pages 30–43, Potsdam, Germany, 2009.Springer-Verlag.

[17] A. Bikakis, T. Patkos, G. Antoniou, and D. Plexousakis. ASurvey of Semantics-based Approaches for Context Reason-ing in Ambient Intelligence. In Proceedings of the Work-shop Artificial Intelligence Methods for Ambient Intelligence,pages 15–24, 2007.

[18] B. Bishop, A. Kiryakov, D. Ognyanoff, I. Peikov, Z. Tashev,and R. Velkov. OWLIM: A family of scalable semantic repos-itories. Semant. web, 2(1):33–42, Jan. 2011.

[19] C. Bizer, T. Heath, and T. Berners-Lee. Linked Data - TheStory So Far. International Journal on Semantic Web andInformation Systems (IJSWIS), 5(3):1–22, MarMar 2009.

[20] F. Bobillo and U. Straccia. Fuzzy ontology representationusing OWL 2. Int. J. Approx. Reasoning, 52(7):1073–1094,2011.

[21] J. Bohn, F. C. Gartner, and H. Vogt. Dependability Issues ofPervasive Computing in a Healthcare Environment. In firstInternational Conference on Security in Pervasive Comput-ing, Boppard, Germany, 2003.

[22] P. A. Bonatti, A. Hogan, A. Polleres, and L. Sauro. Robustand scalable linked data reasoning incorporating provenanceand trust annotations. J. Web Sem., 9(2):165–201, 2011.

[23] S. Borgo, N. Guarino, and C. Masolo. A pointless theory ofspace based on strong connection and congruence. In L. C.Aiello, J. Doyle, and S. Shapiro, editors, KR’96: Principlesof Knowledge Representation and Reasoning, pages 220–229.Morgan Kaufmann, San Francisco, California, 1996.

[24] G. Borriello, V. Stanford, C. Narayanaswami, and W. Men-ning. Guest editors’ introduction: Pervasive computing inhealthcare. IEEE Pervasive Computing, 6(1):17–19, Jan.2007.

[25] B. Bouchard and A. Bouzouane. A key hole plan recognitionmodel for alzheimer’s patients: First results. Applied ArtificialIntelligence, 22:1–34, 2007.

[26] A. Bulling, J. A. Ward, H. Gellersen, and G. Triöster. Ro-bust recognition of reading activity in transit using wearableelectrooculography. In Proceedings of Pervasive 2008, pages19–37. Springer Berlin / Heidelberg, 2008.

[27] P. Buneman, S. Khanna, and W. C.Tan. Data provenance:Some basic issues. In Proceedings of the 20th Conference on

Foundations of Software Technology and Theoretical Com-puter Science, pages 87–93, London, UK, 2000.

[28] M. Burstein, J. Hobbs, O. Lassila, D. Martin, S. McIl-raith, S. Narayanan, M. Paolucci, T. Payne, K. Sycara,and H. Zeng. DAML-S Draft Release (May 2001).http://www.daml.org/services/daml-s/2001/05/, 2001.

[29] M. H. Burstein, J. R. Hobbs, O. Lassila, D. Martin, D. V. Mc-Dermott, S. A. McIlraith, S. Narayanan, M. Paolucci, T. R.Payne, and K. P. Sycara. DAML-S: Web Service Descriptionfor the Semantic Web. In Proceedings of the First Interna-tional Semantic Web Conference on The Semantic Web, ISWC’02, pages 348–363, London, UK, 2002. Springer-Verlag.

[30] R. N. Carvalho, K. B. Laskey, and P. C. G. da Costa. PR-OWL2.0 - Bridging the gap to OWL semantics. In URSW, pages73–84, 2010.

[31] G. Castellano, S. D. Villalba, and A. Camurri. Recognisinghuman emotions from body movement and gesture dynamics.In ACII ’07: Proceedings of the 2nd international conferenceon Affective Computing and Intelligent Interaction, pages 71–82, Lisbon, Portugal, 2007. Springer-Verlag, Berlin, Heidel-berg.

[32] H. Chen, T. Finin, and A. Joshi. An Ontology for Context-Aware Pervasive Computing Environments. Special Issue onOntologies for Distributed Systems, Knowledge EngineeringReview, 18(3):197–207, May 2003.

[33] H. Chen, T. Finin, and A. Joshi. A Pervasive ComputingOntology for User Privacy Protection in the Context BrokerArchitecture. Technical Report TR-CS-04-08, University ofMaryland, Baltimore County, July 2004.

[34] H. Chen, T. Finin, and A. Joshi. Semantic web in the contextbroker architecture. In PERCOM ’04: Proceedings of the Sec-ond IEEE International Conference on Pervasive Computingand Communications (PerCom’04), pages 277 – 286, Wash-ington, DC, USA, 2004. IEEE Computer Society.

[35] H. Chen, T. Finin, and A. Joshi. Ontologies for Agents: The-ory and Experiences, chapter of The SOUPA Ontology forPervasive Computing. Whitestein Series in Software AgentTechnologies. Springer, 2005.

[36] H. Chen, F. Perich, D. Chakraborty, T. Finin, and A. Joshi. In-telligent agents meet semantic web in a smart meeting room.In Proceedings of the Third International Joint Conference onAutonomous Agents and Multiagent Systems - Volume 2, AA-MAS ’04, pages 854–861, New York, New York, 2004. IEEEComputer Society.

[37] H. Chen, F. Perich, T. Finin, and A. Joshi. SOUPA: StandardOntology for Ubiquitous and Pervasive Applications. In FirstAnnual International Conference on Mobile and UbiquitousSystems, Boston, MA, USA, August 2004.

[38] H. L. Chen. An Intelligent Broker Architecture for pervasivecontext-aware systems. PhD thesis, University of Maryland,2004.

[39] L. Chen, C. Nugent, M. Mulvenna, D. Finlay, and X. Hong.Semantic smart homes: Towards knowledge rich assisted liv-ing environments. Intelligent Patient Management, 189:279–296, 2009.

[40] L. Chen and C. D. Nugent. Ontology-based activity recogni-tion in intelligent pervasive environments. International Jour-nal of Web Information Systems, 5(4):410–430, 2009.

[41] L. Chen, C. D. Nugent, and H. Wang. A knowledge-drivenapproach to activity recognition in smart homes. IEEE Trans-actions on Knowledge and Data Engineering, 2011. To ap-


pear.[42] D. Choujaa and N. Dulay. Activity inference through se-

quence alignment. In LoCA ’09: Proceedings of the 4th In-ternational Symposium on location and Context Awareness,pages 19–36, Tokyo, Japan, 2009. Springer-Verlag.

[43] E. Christopoulou, C. Goumopoulos, I. Zaharakis, andA. Kameas. An ontology-based conceptual model for com-posing context-aware applications. In Proceedings of the FirstInternational Workshop on Advanced Context Modelling,Reasoning And Management (in UbiComp 2004), September2004.

[44] X. Chu and R. Buyya. Service oriented sensor web. In SensorNetworks and Configuration, pages 51–74. Springer BerlinHeidelberg, 2007.

[45] A. Ciaramella, M. G. C. A. Cimino, F. Marcelloni, andU. Straccia. Combining fuzzy logic and semantic web to en-able situation-awareness in service recommendation. In Pro-ceedings of the 21st international conference on Databaseand expert systems applications: Part I, DEXA’10, pages 31–45, Berlin, Heidelberg, 2010. Springer-Verlag.

[46] A. G. Cohn, B. Bennett, J. Gooday, and N. M. Gotts. Qualita-tive spatial representation and reasoning with the region con-nection calculus. Geoinformatica, 1(3):275–316, Oct. 1997.

[47] M. Compton, P. Barnaghi, L. Bermudez, R. Garcia-Castro,O. Corcho, S. Cox, J. Graybeal, M. Hauswirth, C. Henson,A. Herzog, V. Huang, K. Janowicz, W. D. Kelsey, D. L.Phuoc, L. Lefort, M. Leggieri, H. Neuhaus, A. Nikolov,K. Page, A. Passant, A. Sheth, and K. Taylor. The SSN Ontol-ogy of the W3C Semantic Sensor Network Incubator Group.Web Semantics: Science, Services and Agents on the WorldWide Web, 0(0), 2012.

[48] M. Compton, C. Henson, L. Lefort, H. Neuhaus, andA. Sheth. A survey of the semantic specification of sensors. InProceedings of the 2nd International Workshop on SemanticSensor Networks (SSN09), pages 17–32, 2009.

[49] D. J. Cook. How smart is your home? Science, 335:1579–1581, Mar. 2012.

[50] D. J. Cook, J. C. Augusto, and V. R. Jakkula. Ambient intelli-gence: Technologies, applications, and opportunities. Perva-sive and Mobile Computing, 5(4):277–298, August 2009.

[51] D. J. Cook and S. K. Das. How smart are our environments?an updated look at the state of the art. Pervasive Mobile Com-puting, 3(2):53–73, 2007.

[52] P. C. Costa, K. B. Laskey, and K. J. Laskey. Uncertainty rea-soning for the Semantic Web I. chapter PR-OWL: A BayesianOntology Language for the Semantic Web, pages 88–107.Springer-Verlag, Berlin, Heidelberg, 2008.

[53] R. Cunningham and V. Cahill. System support for smart cars:requirements and research directions. In Proceedings of the9th workshop on ACM SIGOPS European workshop: beyondthe PC: new challenges for the operating system, EW 9, pages159–164, Kolding, Denmark, 2000. ACM.

[54] S. Dasiopoulou and I. Kompatsiaris. Trends and issues in de-scription logics frameworks for image interpretation. In Pro-ceedings of the 6th Hellenic conference on Artificial Intel-ligence: theories, models and applications, SETN’10, pages61–70, Athens, Greece, 2010. Springer-Verlag.

[55] Dublin Core Metadata Initiative. http://dublincore.org.

[56] S. Decker, M. Erdmann, D. Fensel, and R. Studer. Ontobro-ker: Ontology based access to distributed and semi-structured

information. In R. Meersman, Z. Tari, and S. M. Stevens,editors, DS-8, volume 138 of IFIP Conference Proceedings,pages 351–369. Kluwer, 1999.

[57] A. K. Dey. Understanding and Using Context. Personal andUbiquitous Computing, 5(1):4–7, 2001.

[58] Z. Ding, Y. Peng, R. Pan, Z. Ding, Y. Peng, and R. Pan.BayesOWL: Uncertainty modeling in semantic web ontolo-gies. Soft Computing in Ontologies and Semantic Web, pages3–29, 2006.

[59] R. Dividino, S. Sizov, S. Staab, and B. Schueler. Queryingfor provenance, trust, uncertainty and other meta knowledgein rdf. Web Semant., 7:204–219, September 2009.

[60] S. Dobson. Leveraging the subtleties of location. In Proceed-ings of the 2005 joint conference on Smart objects and am-bient intelligence: innovative context-aware services: usagesand technologies, sOc-EUSAI ’05, pages 189–193, Grenoble,France, 2005. ACM.

[61] M. Doerr. The CIDOC conceptual reference module: an onto-logical approach to semantic interoperability of metadata. AIMag., 24(3):75–92, Sept. 2003.

[62] dotNetRDF - Semantic Web/RDF Library for C#/.Net.http://www.dotnetrdf.org/.

[63] J. Du, G. Qi, Y.-D. Shen, and J. Z. Pan. Towards practicalABox abduction in large OWL DL ontologies. In Proceed-ings of the Twenty-Fifth AAAI Conference on Artificial Intel-ligence, AAAI 2011, pages 1160–1165, San Francisco, Cali-fornia, USA, 2011. AAAI Press.

[64] DOLCE+DnS Ultralite (DUL) ontology. http://www.loa.istc.cnr.it/ontologies/DUL.owl.

[65] E. Dumbill. Finding friends with XMLand RDF: FOAF, 2002. http://www-106.ibm.com/developerworks/xml/library/x-foaf.html.

[66] V. Eisenberg and Y. Kanza. D2rq/update: updating relationaldata via virtual rdf. In Proceedings of the 21st internationalconference companion on World Wide Web, WWW ’12 Com-panion, pages 497–498, Lyon, France, 2012. ACM.

[67] R. Elmasri, G. T. J. Wuu, and Y.-J. Kim. The time index—an access structure for temporal data. In Proceedings of thesixteenth international conference on Very large databases,pages 1–12, Brisbane, Australia, 1990. Morgan KaufmannPublishers Inc.

[68] O. Erling. Virtuoso, a Hybrid RDBMS/Graph Column Store.IEEE Data Eng. Bull., 35(1):3–8, 2012.

[69] K. Farkas, H.-P. Labs, J. Heidemann, and L. Iftode. Intelli-gent transportation and pervasive computing. Pervasive Com-puting, 5:18–19, 2006.

[70] G. Fenza, V. Loia, and S. Senatore. A hybrid approach to se-mantic web services matchmaking. Int. J. Approx. Reasoning,48(3):808–828, 2008.

[71] Davies and Vitiello’s relationship vocabulary for FOAF.http://vocab.org/relationship.

[72] D. Franklin. Cooperating with people: the intelligent class-room. In Proceedings of the fifteenth national/tenth con-ference on Artificial intelligence/Innovative applications ofartificial intelligence, AAAI ’98/IAAI ’98, pages 555–560,Madison, Wisconsin, United States, 1998. American Associ-ation for Artificial Intelligence.

[73] A. Gangemi and P. Mika. Understanding the semantic webthrough descriptions and situations. In Proceedings of the In-ternational Conference on Ontologies, Databases and Appli-cations of SEmantics, pages 689–706, 2003.


[74] J. Gómez-Romero, M. A. Patricio, J. García, and J. M.Molina. Context-based reasoning using ontologies to adaptvisual tracking in surveillance. In Proceedings of the 2009Sixth IEEE International Conference on Advanced Video andSignal Based Surveillance, AVSS ’09, pages 226–231, Wash-ington, DC, USA, 2009. IEEE Computer Society.

[75] J. Gómez-Romero, M. A. Patricio, J. García, and J. M.Molina. Ontology-based context representation and reasoningfor object tracking and scene interpretation in video. ExpertSyst. Appl., 38(6):7494–7510, 2011.

[76] C. Goumopoulos, A. D. Kameas, and A. Cassells. Anontology-driven system architecture for precision agricultureapplications. IJMSO, 4(1/2):72–84, 2009.

[77] B. C. Grau, I. Horrocks, B. Motik, B. Parsia, P. Patel-Schneider, and U. Sattler. OWL 2: The Next Step for OWL.Web Semantics: Science, Services and Agents on the WorldWide Web, 6(4):309–322, October 2008.

[78] B. N. Grosof, I. Horrocks, R. Volz, and S. Decker. Descriptionlogic programs: combining logic programs with descriptionlogic. In WWW, pages 48–57, 2003.

[79] T. Gu, H. K. Pung, and D. Q. Zhang. A service-oriented mid-dleware for building context-aware services. J. Netw. Comput.Appl., 28(1):1–18, Jan. 2005.

[80] T. Gu, X. H. Wang, H. K. Pung, and D. Q. Zhang.An Ontology-based Context Model in Intelligent Environ-ments. In Proceedings of the Communication Networks andDistributed Systems Modeling and Simulation Conference(CNDS 2004), pages 270–275, January 2004.

[81] C. Gutierrez, C. A. Hurtado, and A. Vaisman. Introducingtime into rdf. IEEE Transactions on Knowledge and DataEngineering, 19(2):207–218, 2007.

[82] V. Haarslev and R. Möller. Racer: A Core Inference En-gine for the Semantic Web. In Proceedings of the 2ndInternational Workshop on Evaluation of Ontology-basedTools (EON2003), located at the 2nd International SemanticWeb Conference, Sanibel Island, Florida, USA, pages 27–36,2003.

[83] J. K. Hart and K. Martinez. Environmental sensor networks:arevolution in the earth system science? Earth-Science Re-views, 78:177–191, 2006.

[84] K. Henricksen and J. Indulska. Modelling and using imper-fect context information. In PerCom Workshops, pages 33–37, 2004.

[85] M. Hepp. Goodrelations: An ontology for describing productsand services offers on the web. In Proceedings of the 16th in-ternational conference on Knowledge Engineering: Practiceand Patterns, EKAW ’08, pages 329–346, Berlin, Heidelberg,2008. Springer-Verlag.

[86] M. Hert, G. Reif, and H. C. Gall. A comparison of RDB-to-RDF mapping languages. In Proceedings of the 7th Inter-national Conference on Semantic Systems, I-Semantics ’11,pages 25–32, Graz, Austria, 2011. ACM.

[87] J. Hightower, B. Brumitt, and G. Borriello. The locationstack: A layered model for location in ubiquitous computing.Proceedings of WMCSA 2002, 00:22–28, 2002.

[88] J. R. Hobbs and F. Pan. An ontology of time for the seman-tic web. ACM Transactions on Asian Language InformationProcessing (TALIP), 3(1):66–85, 2004.

[89] F. Hohl, U. Kubach, A. Leonhardi, K. Rothermel, andM. Schwehm. Next century challenges: Nexus—an openglobal infrastructure for spatial-aware applications. In Mo-

biCom ’99: Proceedings of the 5th annual ACM/IEEE inter-national conference on Mobile computing and networking,pages 249–255, New York, NY, USA, 1999. ACM.

[90] M. Horridge and S. Bechhofer. The OWL API: A Java APIfor OWL ontologies. Semant. web, 2(1):11–21, Jan. 2011.

[91] I. Horrocks and P. F. Patel-Schneider. KR and reasoning onthe semantic web: OWL. In J. Domingue, D. Fensel, and J. A.Hendler, editors, Handbook of Semantic Web Technologies,chapter 9, pages 365–398. Springer, 2011.

[92] I. Horrocks, P. F. Patel-Schneider, H. Boley, S. Tabet,B. Grosof, and M. Dean. SWRL: A Semantic Web Rule Lan-guage Combining OWL and RuleML. Technical report, Na-tional Research Council of Canada, Network Inference, andStanford University, May 2004.

[93] I. Horrocks, P. F. Patel-Schneider, and F. van Harmelen. FromSHIQ and RDF to OWL: the making of a Web Ontology Lan-guage. J. Web Sem., 1(1):7–26, 2003.

[94] S. Hossein, S. Hedal, and A. Mendez-Vasquez. Sensorydata set description language specification. Technical ReportSDDL_Specification_v1.0, University of Florida, 2009.

[95] International Standards Organisation (ISO). ISO-8601: Dataelements and interchange formats – Information interchange– Representation of dates and times. ISO, Geneva, Switzer-land, 2004.

[96] International organization for standardization, for FDIS19115 geographic information – metadata. http://www.ncits.org/ref-docs/FDIS_19115.pdf.

[97] V. R. Jakkula, A. S. Crandall, and D. J. Cook. Knowledgediscovery in entity based smart environment resident data us-ing temporal relation based data mining. In ICDMW ’07:Proceedings of the Seventh IEEE International Conference onData Mining Workshops, pages 625–630, Washington, DC,USA, 2007. IEEE Computer Society.

[98] K. Janowicz and M. Compton. The stimulus-sensor-observation ontology design pattern and its integration intothe semantic sensor network ontology. In Proceedings ofthe 3rd International workshop on Semantic Sensor Networks2010 (SSN10) in conjunction with the 9th International Se-mantic Web Conference (ISWC 2010), Shanghai, China, Sept.2010.

[99] Apache Jena. http://jena.apache.org/.[100] C. Jiang and P. Steenkiste. A hybrid location model with a

computable location identifier for ubiquitous computing. InProceedings of UbiComp ’02, pages 246–263, Gothenberg,Sweden, 2002. Springer-Verlag.

[101] L. Kagal, T. Finin, and A. Joshi. A policy based approach tosecurity for the semantic web. In Proceedings of the secondInternational Semantic Web Conference (ISWC 2003), pages402–418, Sanibel Island, Florida, USA, 2003.

[102] L. Kagal, V. Korolev, H. Chen, A. Joshi, and T. Finin. Centau-rus: A framework for intelligent services in a mobile environ-ment. In ICDCSW ’01: Proceedings of the 21st InternationalConference on Distributed Computing Systems, pages 195–201, Washington, DC, USA, April 2001. IEEE Computer So-ciety.

[103] B. M. Kiernan, S. Beirne, C. Fay, and D. Diamond. Mon-itoring of gas emissions at landfill sites using autonomousgas sensors. Technical report, Clarity Centre, Ireland, 2010.Project Report. STRIVE, Environmental Protection Agency.

[104] M. Kifer, G. Lausen, and J. Wu. Logical foundations ofobject-oriented and frame-based languages. Journal of ACM,


42(4):741–843, 1995.[105] S. Klarman, U. Endriss, and S. Schlobach. ABox Abduc-

tion in the Description Logic ALC. J. Autom. Reasoning,46(1):43–80, 2011.

[106] M. Klusch. Chapter 4: Semantic web service coordination. InCASCOM - Intelligent Service Coordination in the SemanticWeb. Birkhaeuser Verlag, Springer, 2008.

[107] R. A. Kowalski and M. J. Sergot. A logic-based calculus ofevents. New Generation Comput., 4(1):67–95, 1986.

[108] A. A. Krisnadhi, K. Sengupta, and P. Hitzler. Local closedworld semantics: Keep it simple, stupid! In Description Log-ics, 2011.

[109] K. N. Kumar, R.Prabakaran, V. S. Dhulipala, and P.Ranjith.Future sensors and utilization of sensors in chemical indus-tries with control of environmental hazards. In Proceedings ofthe 2nd International Conference on Environmental Scienceand Development, pages 224–228. IACSIT Press, 2011.

[110] U. Küster, H. Lausen, and B. König-Ries. Evaluation of se-mantic service discovery – a survey and directions for fu-ture research. In Post-Proceedings of the 2nd Workshop onEmerging Web Services Technology (WEWST07) in conjunc-tion with the 5th IEEE European Conference on Web Services(ECOWS07), 2007.

[111] D. B. Lenat and R. V.Guha. Building Large Knowledge-BasedSystems; Representation and Inference in the Cyc Project.Addison-Wesley Longman Publishing Co., Inc., Boston, MA,USA, 1989.

[112] L. Li and I. Horrocks. A software framework for matchmak-ing based on semantic web technology. In Proceedings ofthe 12th international conference on World Wide Web, WWW’03, pages 331–339, Budapest, Hungary, 2003. ACM.

[113] P. Lilis, I. Lourdi, C. Papatheodorou, and M. Gergatsoulis. Ametadata model for representing time-dependent informationin cultural collections. In Proceedings of the first online meta-data and semantics research conference (MTSR’ 05), pages1–12. Rinton Press, Nov. 2005.

[114] B. Logan, J. Healey, M. Philipose, E. M. Tapia, and S. S. In-tille. A long-term evaluation of sensing modalities for activityrecognition. In UbiComp’07: Proceedings of the 9th interna-tional conference on Ubiquitous computing, pages 483–500,Innsbruck, Austria, September 2007. Springer-Verlag.

[115] D. Martin, M. Burstein, and J. Hobbs.OWL-S: Semantic markup for web services.http://www.w3.org/Submission/OWL-S, 2004.

[116] K. Martinez, J. K. Hart, and R. Ong. Environmental sensornetworks. Computer, 37(8):50–56, Aug. 2004.

[117] B. McBride. Jena: A semantic web toolkit. IEEE InternetComputing, 6(6):55–59, 2002.

[118] B. McBride and M. Butler. Representing and querying his-torical information in RDF with application to E-discovery.Technical Report HPL-2009-261, Hewlit Packard Laborato-ries, 2009.

[119] I. McCowan, D. Gatica-Perez, S. Bengio, G. Lathoud,M. Barnard, and D. Zhang. Automatic analysis of multimodalgroup actions in meetings. IEEE Transactions on PatternAnalysis and Machine Intelligence, 27:305–317, 2005.

[120] R. E. McGrath. Discovery and Its Discontents: Discov-ery Protocols for Ubiquitous Computing. Technical Re-port UIUCDCS-R-99-2132, Department of Computer Sci-ence University of Illinois Urbana-Champaign , Urbana,March 2000.

[121] R. E. McGrath, A. Ranganathan, R. H. Campbell, and M. D.Mickunas. Use of Ontologies in a Pervasive Computing Envi-ronment. Technical Report UIUCDCS-R-2003-2332 UILU-ENG-2003-1719, Department of Computer Science, Univer-sity of Illinois, Urbana-Champaign, Urbana, Illinois, 2003.

[122] R. E. McGrath, A. Ranganathan, M. D. Mickunas, and R. H.Campbell. Investigations of semantic interoperability inubiquitous computing environments. In Proceedings of the15th IASTED International Conference on Parallel And Dis-tributed Computing And Systems (PDCS 2003), Marina delRey, CA, USA, November 2003.

[123] S. McKeever, J. Ye, L. Coyle, and S. Dobson. A multilayereduncertainty model for context aware systems. In Adjunct Pro-ceedings of the international conference on Pervasive Com-puting: Late Breaking Result, pages 1–4, Sydney, Australia,May 2008.

[124] G. Meditskos and N. Bassiliades. A Rule-Based Object-Oriented OWL Reasoner. IEEE Trans. Knowl. Data Eng.,20(3):397–410, 2008.

[125] G. Meditskos and N. Bassiliades. DLEJena: A practicalforward-chaining OWL 2 RL reasoner combining Jena andPellet. Web Semant., 8(1):89–94, Mar. 2010.

[126] S. B. Mokhtar, D. Preuveneers, N. Georgantas, V. Issarny, andY. Berbers. Easy: Efficient semantic service discovery in per-vasive computing environments with qos and context support.Journal of Systems and Software, 81(5):785 – 808, 2008.

[127] L. Moreau, B. Clifford, J. Freire, J. Futrelle, Y. Gil, P. Groth,N. Kwasnikowska, S. Miles, P. Missier, J. Myers, B. Plale,Y. Simmhan, E. Stephan, and J. V. den Bussche. The OpenProvenance Model core specification (v1.1). Future Genera-tion Computer Systems, July 2010.

[128] T. Moser, H. Roth, S. Rozsnyai, R. Mordinyi, and S. Biffl. Se-mantic event correlation using ontologies. In Proceedings ofthe Confederated International Conferences, CoopIS, DOA,IS, and ODBASE 2009 on On the Move to Meaningful Inter-net Systems: Part II, OTM ’09, pages 1087–1094, Vilamoura,Portugal, 2009. Springer-Verlag.

[129] B. Motik, I. Horrocks, R. Rosati, and U. Sattler. Can OWLand logic programming live together happily ever after? In In-ternational Semantic Web Conference, pages 501–514, 2006.

[130] B. Motik and R. Rosati. Reconciling description logics andrules. J. ACM, 57(5):30:1–30:62, June 2008.

[131] B. Motik, U. Sattler, and R. Studer. Query Answering forOWL-DL with rules. J. Web Sem., 3(1):41–60, 2005.

[132] B. Motik, R. Shearer, and I. Horrocks. Hypertableau reason-ing for description logics. J. Artif. Int. Res., 36(1):165–228,Sept. 2009.

[133] M. Mühlenbrock, O. Brdiczka, D. Snowdon, and J.-L. Meu-nier. Learning to detect user activity and availability from avariety of sensor data. In PERCOM ’04: Proceedings of theSecond IEEE International Conference on Pervasive Comput-ing and Communications, pages 13 – 22, Orlando, Florida,Mar. 2004. IEEE Computer Society.

[134] The NeOn Toolkit. http://neon-toolkit.org.[135] R. Nevatia, J. Hobbs, and B. Bolles. An ontology for video

event representation. In Proceedings of the 2004 Confer-ence on Computer Vision and Pattern Recognition Workshop(CVPRW’04) Volume 7, CVPRW ’04, pages 119–129, Wash-ington, DC, USA, 2004. IEEE Computer Society.

[136] N. F. Noy, R. W. Fergerson, and M. A. Musen. The knowl-edge model of Protége-2000: Combining interoperability and


flexibility. In Proceedings of the 2th International Confer-ence on Knowledge Engineering and Knowledge Manage-ment (EKAW’ 2000), pages 17–32, Juan-les-Pins, France,2000. Springer.

[137] M. J. O’Connor and A. K. Das. SQWRL: A query lan-guage for OWL. In OWL: Experiences and Directions(OWLED), volume 529 of CEUR Workshop Proceedings.CEUR-WS.org, 2009.

[138] M. J. O’Connor and A. K. Das. A lightweight modelfor representing and reasoning with temporal information inbiomedical ontologies. In Proceedings of the InternationalConference on Health Informatics(HEALTHINF 2010), Va-lencia, Spain, 2010.

[139] F. Orlandi and A. Passant. Modelling provenance of dbpe-dia resources using wikipedia contributions. J. Web Sem.,9(2):149–164, 2011.

[140] F. Paganelli and D. Giuli. An ontology-based systemfor context-aware and configurable services to supporthome-based continuous care. Information Technology inBiomedicine, IEEE Transactions on, 15(2):324 –333, March2011.

[141] F. Pan and J. R. Hobbs. Time ontology inOWL. W3C working draft, W3C, Sept. 2006.http://www.w3.org/TR/2006/WD-owl-time-20060927/.

[142] J. Paradiso, P. Dutta, H. Gellersen, and E. Schooler. Guesteditors’ introduction: Smart energy systems. Pervasive Com-puting, IEEE, 10(1):11 –12, Jan-Mar 2011.

[143] A. Paschke, H. Boley, and P. Vincent. Semantic complexevent processing – the future of dynamic IT. Invited talk atSemTech 2010, June 2010.

[144] P. F. Patel-Schneider and I. Horrocks. A comparison of twomodelling paradigms in the semantic web. J. Web Sem.,5(4):240–250, 2007.

[145] T. Patkos, I. Chrysakis, A. Bikakis, D. Plexousakis, andG. Antoniou. A reasoning framework for ambient intel-ligence. In Proceedings of the 6th Hellenic conferenceon Artificial Intelligence: theories, models and applications,SETN’10, pages 213–222, Athens, Greece, 2010. Springer-Verlag.

[146] I. S. E. Peraldi, A. Kaya, and R. Möller. Formalizing multime-dia interpretation based on abduction over description logicABoxes. In Description Logics, volume 477 of CEUR Work-shop Proceedings, 2009.

[147] F. Perich, A. Joshi, T. Finin, and Y. Yesha. On data manage-ment in pervasive computing environments. IEEE Trans. onKnowl. and Data Eng., 16(5):621–634, May 2004.

[148] M. Perry, A. Sheth, and P. Jain. SPARQLST:ExtendingSPARQL to Support Spatiotemporal Queries. Technical Re-port KNOESIS-TR-2009-01, Kno.e.sis Center Technical Re-port, 2009.

[149] PROV-O: The PROV Ontology, W3C Working Draft 03 May2012. http://www.w3.org/TR/prov-o/.

[150] Provenance Vocabulary Core Ontology Specification, 14March 2012. http://trdf.sourceforge.net/provenance/ns.html.

[151] Provenance Working Group. http://www.w3.org/2011/prov/wiki/Main_Page.

[152] Provenir Ontology. http://wiki.knoesis.org/index.php/Provenir_Ontology.

[153] A. Pugliese, O. Udrea, and V. S. Subrahmanian. Scaling RDFwith Time. In Proceeding of the 17th international confer-

ence on World Wide Web (WWW ’08), pages 605–614, Bei-jing, China, 2008. ACM.

[154] A. Ranganathan, J. Al-Muhtadi, and R. H. Campbel. Reason-ing about uncertain contexts in pervasive computing environ-ments. IEEE Pervasive Computing, 3(2):62–70, 2004.

[155] A. Ranganathan, J. Al-Muhtadi, S. Chetan, R. Campbell, andM. D. Mickunas. Middlewhere: a middleware for locationawareness in ubiquitous computing applications. In Proceed-ings of Middleware ’04, pages 397–416, New York, USA,2004. Springer-Verlag New York, Inc.

[156] A. Ranganathan and R. H. Campbell. An infrastructure forcontext-awareness based on first order logic. Personal andUbiquitous Computing, 7:353–364, 2003.

[157] A. Ranganathan, R. E. McGrath, R. H. Campbell, and M. D.Mickunas. Use of Ontologies in a Pervasive Computing En-vironment. Knowledge Engineering Review, 18(3):209–220,2004.

[158] A. S. Rao and M. P. Georgeff. BDI agents: From theory topractice. In ICMAS, pages 312–319. The MIT Press, 1995.

[159] D. Riboni and C. Bettini. Cosar: hybrid reasoning for context-aware activity recognition. Personal and Ubiquitous Comput-ing, 15(3):271–289, 2011.

[160] D. Riboni and C. Bettini. OWL 2 modeling and reasoningwith complex human activities. Pervasive and Mobile Com-puting, 7(3):379–395, 2011.

[161] M. Roman, C. K. Hess, R. Cerqueira, A. Ranganathan, R. H.Campbell, and K. Nahrstedt. Gaia: A Middleware Infrastruc-ture to Enable Active Spaces. IEEE Pervasive Computing,pages 74–83, Oct–Dec 2002.

[162] R. Rosati. DL+log: Tight integration of description logicsand disjunctive datalog. In P. Doherty, J. Mylopoulos, andC. A. Welty, editors, Proceedings of the tenth InternationalConference on Principles of Knowledge Representation andReasoning, pages 68–78. AAAI Press, 2006.

[163] S. Rozsnyai, R. Vecera, J. Schiefer, and A. Schatten. Eventcloud - searching for correlated business events. In IEEE In-ternational Conference on E-Commerce Technology and En-terprise Computing, E-Commerce, and E-Services, volume 0,pages 409–420, Los Alamitos, CA, USA, 2007. IEEE Com-puter Society.

[164] D. Russomanno, C. Kothari, and O. Thomas. Sensor ontolo-gies: From shallow to deep models. In Proceedings of the37th Southeastern Symposium on System Theory (SSST ’05),pages 107–112, Mar. 2005.

[165] D. J. Russomanno, C. R. Kothari, and O. A. Thomas. Build-ing a sensor ontology: A practical approach leveraging ISOand OGC models. In Proceedings of the 2005 InternationalConference on Artificial Intelligence, pages 637–643, Las Ve-gas, USA, 2005.

[166] K. Sagonas, T. Swift, and D. S. Warren. Xsb as an efficient de-ductive database engine. In In Proceedings of the ACM SIG-MOD International Conferenceon the Management of Data,pages 442–453. ACM Press, 1994.

[167] S. Santini, B. Ostermaier, and A. Vitaletti. First experiencesusing wireless sensor networks for noise pollution monitor-ing. In Proceedings of the workshop on Real-world wire-less sensor networks, REALWSN ’08, pages 61–65, Glasgow,Scotland, 2008. ACM.

[168] A. Scherp, T. Franz, C. Saathoff, and S. Staab. A core ontol-ogy on events for representing occurrences in the real world.Multimedia Tools and Applications, 58(2):293–331, 2012.


[169] A. Scherp, S. Papadopoulos, A. Kritikos, F. Schwa-gereit, C. Saathoff, T. Franz, D. Schmeiss, S. Staab,S. Schenk, and M. Bonifacio. Prototypical Knowl-edge Management Methodology (Deliverable D5.2.1), We-KnowIt Project. http://www.weknowit.eu/sites/default/files/D5.2.1.pdf, 2009.

[170] H. R. Schmidtke and W. Woo. Towards ontology-based for-mal verification methods for context aware systems. In Per-vasive ’09: Proceedings of the 7th international conferenceon pervasive computing, pages 309–326, Nara, Japan, 2009.Springer-Verlag.

[171] openRDF.org. http://www.openrdf.org/.[172] M. Shanahan. Perception as abduction: Turning sensor

data into meaningful representation. Cognitive Science,29(1):103–134, 2005.

[173] R. Shaw, R. Troncy, and L. Hardman. Lode: Linking opendescriptions of events. In Proceedings of the 4th Asian Con-ference on The Semantic Web, ASWC ’09, pages 153–167,Shanghai, China, 2009. Springer-Verlag.

[174] E. Sirin, B. Parsia, B. C. Grau, A. Kalyanpur, and Y. Katz.Pellet: A practical OWL-DL reasoner. Web Semantics: Sci-ence, Services and Agents on the World Wide Web, 5(2):51–53, 2011.

[175] S. Song, M. Kim, S. Rho, and E. Hwang. Music ontology formood and situation reasoning to support music retrieval andrecommendation. In Proceedings of the 2009 Third Interna-tional Conference on Digital Society, ICDS ’09, pages 304–309, Washington, DC, USA, 2009. IEEE Computer Society.

[176] T. Springer and A.-Y. Turhan. Employing description log-ics in ambient intelligence for modeling and reasoning aboutcomplex situations. JAISE, 1(3):235–259, 2009.

[177] G. Stevenson, S. Knox, S. Dobson, and P. Nixon. Ontonym:a collection of upper ontologies for developing pervasive sys-tems. In Proceedings of the 1st Workshop on Context, Infor-mation and Ontologies, CIAO ’09, pages 9:1–9:8, Heraklion,Greece, 2009. ACM.

[178] G. Stevenson, J. Ye, and S. Dobson. On the impact of thetemporal features of sensed data on the development of per-vasive systems. In Proceedings of the International Workshopon Programming Methods for Mobile and Pervasive Systems(PMMPS’ 10), Helsinki, Finland, May 2010.

[179] G. Stevenson, J. Ye, S. Dobson, and P. Nixon. Loc8: A loca-tion model and extensible framework for programming withlocation. IEEE Pervasive Computing, 9(1):28 – 37, January2010.

[180] G. Stoilos, G. B. Stamou, and J. Z. Pan. Fuzzy extensionsof owl: Logical properties and reduction to fuzzy descriptionlogics. Int. J. Approx. Reasoning, 51(6):656–679, 2010.

[181] G. Stoilos, G. B. Stamou, J. Z. Pan, V. Tzouvaras, and I. Hor-rocks. Reasoning with very expressive fuzzy description log-ics. J. Artif. Intell. Res. (JAIR), 30:273–320, 2007.

[182] U. Straccia. Towards a fuzzy description logic for the se-mantic web (preliminary report). In Proceedings of the Sec-ond European conference on The Semantic Web: research andApplications, ESWC’05, pages 167–181, Heraklion, Greece,2005. Springer-Verlag.

[183] U. Straccia. Reasoning web. chapter Managing Uncertaintyand Vagueness in Description Logics, Logic Programs andDescription Logic Programs, pages 54–103. Springer-Verlag,Berlin, Heidelberg, 2008.

[184] T. Strang and C. Linnhoff-Popien. Service interoperabil-ity on context level in ubiquitous computing environments.In Proceedings of International Conference on Advances inInfrastructure for Electronic Business, Education, Science,Medicine, and Mobile Technologies on the Internet (SS-GRR2003w), L’Aquila/Italy, January 2003.

[185] T. Strang and C. Linnhoff-Popien. A context modeling sur-vey. In Proceedings of the Workshop on Advanced Con-text Modelling, Reasoning and Management as part of Ubi-Comp 2004 - The Sixth International Conference on Ubiqui-tous Computing, Nottingham/England, September 2004.

[186] T. Strang, C. Linnhoff-Popien, and K. Frank. Applications ofa Context Ontology Language. In D. Begusic and N. Rozic,editors, Proceedings of the International Conference on Soft-ware, Telecommunications and Computer Networks (Soft-Com2003), pages 14–18. Faculty of Electrical Engineering,Mechanical Engineering and Naval Architecture, Universityof Split, Croatia, October 2003.

[187] T. Strang, C. Linnhoff-Popien, and K. Frank. CoOL: A Con-text Ontology Language to enable Contextual Interoperabil-ity. In J.-B. Stefani, I. Dameure, and D. Hagimont, edi-tors, LNCS 2893: Proceedings of 4th IFIP WG 6.1 Interna-tional Conference on Distributed Applications and Interop-erable Systems (DAIS2003), Lecture Notes in Computer Sci-ence (LNCS), pages 236–247, Paris/France, November 2003.Springer Verlag.

[188] N. Streitz and P. Nixon. Introduction. Commun. ACM,48(3):32–35, Mar. 2005.

[189] IEEE Suggested Upper Merged Ontology. http://suo.ieee.org/SUO/SUMO/SUMO_173.kif.

[190] J. Tao, E. Sirin, J. Bao, and D. L. McGuinness. Integrity Con-straints in OWL. In Proceedings of Twenty-Fourth AAAI Con-ference on Artificial, 2010.

[191] J. Tappolet and A. Bernstein. Applied Temporal RDF: Effi-cient Temporal Querying of RDF Data with SPARQL. In Pro-ceedings of the 6th European Semantic Web Conference onThe Semantic Web, pages 308–322, Heraklion, Crete, Greece,2009. Springer-Verlag.

[192] K. Taylor and L. Leidinger. Ontology-driven complex eventprocessing in heterogeneous sensor networks. In Proceedingsof the 8th extended semantic web conference on The semanticweb: research and applications - Volume Part II, pages 285–299, Heraklion, Crete, Greece, 2011. Springer-Verlag.

[193] H. J. ter Horst. Extending the rdfs entailment lemma. InInternational Semantic Web Conference, pages 77–91, 2004.

[194] H. J. ter Horst. Completeness, decidability and complexity ofentailment for rdf schema and a semantic extension involvingthe owl vocabulary. Web Semant., 3(2-3):79–115, Oct. 2005.

[195] K. Teymourian and A. Paschke. Semantic rule-based complexevent processing. In Proceedings of the 2009 InternationalSymposium on Rule Interchange and Applications, RuleML’09, pages 82–92, Las Vegas, Nevada, 2009. Springer-Verlag.

[196] Y. Theoharis, I. Fundulaki, G. Karvounarakis, andV. Christophides. On provenance of queries on semantic webdata. IEEE Internet Computing, 15(1):31–39, Jan. 2011.

[197] A. Toninelli, R. Montanari, L. Kagal, and O. Lassila. A se-mantic context-aware access control framework for securecollaborations in pervasive computing environments. In Pro-ceedings of the 5th international conference on The SemanticWeb, ISWC’06, pages 473–486, Athens, GA, 2006. Springer-Verlag.


[198] TopBraid Composer. http://www.topquadrant.com/products/TB_Composer.html.

[199] D. Trastour, C. Bartolini, and J. Gonzalez-Castillo. A Seman-tic Web Approach to Service Description for Matchmaking ofServices. Technical Report HPL-2001-183, HP LaboratoriesBristol, 2001.

[200] D. Tsarkov and I. Horrocks. FaCT++ Description Logic Rea-soner: System Description. In International Joint Confer-ence on Automated Reasoning (IJCAR 2006), pages 292–297,2006.

[201] A.-Y. Turhan, T. Springer, and M. Berger. Pushing doors formodeling contexts with OWL DL a case study. In PerComWorkshops, pages 13–17, 2006.

[202] M. Vacura, V. Svátek, and P. Smrž. A pattern-based frame-work for uncertainty representation in ontologies. In Proceed-ings of the 11th international conference on Text, Speech andDialogue, TSD ’08, pages 227–234, Brno, Czech Republic,2008. Springer-Verlag.

[203] W. R. van Hage, V. Malaise, R. H. Segers, L. Hollink, andG. Schreiber. Design and use of the simple event model (sem).Web Semantics: Science, Services and Agents on the WorldWide Web, 9(2), 2011.

[204] F. van Harmelen and D. L. McGuinness. OWL Web OntologyLanguage Overview. http://www.w3.org/TR/2004/REC-owl-features-20040210/, 2004.

[205] T. van Kasteren, A. Noulas, G. Englebienne, and B. Kröse.Accurate activity recognition in a home setting. In UbiComp’08: Proceedings of the 10th International Conference onUbiquitous Computing, pages 1–9, Seoul, Korea, Sept. 2008.ACM.

[206] M. Y. Vardi. Why is modal logic so robustly decidable? InProceedings of a DIMACS Workshop on Descriptive Com-plexity and Finite Models, DIMACS Series in Discrete Math-ematics and Theoretical Computer Science, pages 149–184.American Mathematical Society, Jan. 1996.

[207] M. Völkel and Y. Sure. RDFReactor - From Ontologies toProgrammatic Data Access. In Poster and Demo proceedingsfo the International Semantic Web Conference (ISWC) 2005,Galway, Ireland, Nov. 2005.

[208] W3C Semantic Sensor Network Incubator group. http://www.w3.org/2005/Incubator/ssn/.

[209] Review of sensor and observation ontologies.http://www.w3.org/2005/Incubator/ssn/wiki/Review_of_Sensor_and_Observations_Ontologies.

[210] X. Wang, J. S. Dong, C.-Y. Chin, S. Hettiarachchi, andD. Zhang. Semantic Space: An Infrastructure for SmartSpaces. IEEE Pervasive Computing, 3(3):32–39, 2004.

[211] X. Wang, D. Zhang, T. Gu, and H. K. Pung. Ontology basedcontext modeling and reasoning using owl. In PerCom Work-shops, pages 18–22, 2004.

[212] M. Weiser. The computer for the 21st century. ScientificAmerican, 265(3):94–104, Sept. 1991.

[213] K. J. Witt, J. Stanley, D. Smithbauer, D. Mandl, A. U. V. Ly,and M. Metheny. Enabling sensor webs by utilizing SWAMOfor autonomous operations. In Proceedings of the 8th NASAEarth Science Technology Conference (ESTC 2008), 2008.

[214] T. W. Wlodarczyk, C. Rong, M. O’Connor, and M. Musen.SWRL-F: a fuzzy logic extension of the semantic web rulelanguage. In Proceedings of the International Conference onWeb Intelligence, Mining and Semantics, WIMS ’11, pages39:1–39:9, Sogndal, Norway, 2011. ACM.

[215] J. Ye, L. Coyle, S. Dobson, and P. Nixon. Ontology-basedmodels in pervasive computing systems. The Knowledge En-gineering Review, 22(04):315–347, December 2007.

[216] J. Ye, L. Coyle, S. Dobson, and P. Nixon. Using situationlattices to model and reason about context. In MRC 2007:Proceedings of the Workshop on modeling and reasoning oncontext (coexist with CONTEXT’07), pages 1–12, Roskilde,Denmark, August 2007.

[217] J. Ye, S. Dobson, and S. McKeever. Situation identificationtechniques in pervasive computing: a review. Pervasive andmobile computing, 8:36–66, Feb. 2012.

[218] J. Ye, S. McKeever, L. Coyle, S. Neely, and S. Dobson. Re-solving uncertainty in context integration and abstraction. InICPS 2008: Proceedings of the international conference onPervasive Services, pages 131–140, Sorrento, Italy, July 2008.ACM.

[219] J. Ye, G. Stevenson, and S. Dobson. A top-level ontology forsmart environments. Pervasive and mobile computing, 7:359–378, June 2011.

[220] J. Ye, G. Stevenson, S. Dobson, M. O’Grady, and G. O’Hare.PI: Perceiver and interpreter of smart home datasets. In Pro-ceedings of the 5th International ICST Conference on Per-vasive Computing Technologies for Healthcare (Pervasive-Health 2011), pages 131–138, Dublin, Ireland, 2011.

[221] Z. Yu and Y. Nakamura. Smart meeting systems: A sur-vey of state-of-the-art and open issues. ACM Comput. Surv.,42(2):8:1–8:20, Mar. 2010.

[222] Z. Yu, X. Zhou, Z. Yu, J. H. Park, and J. Ma. iMuseum: Ascalable context-aware intelligent museum system. ComputerCommunications, 31(18):4376–4382, 2008.

[223] J. Yuan, Y. Zheng, X. Xie, and G. Sun. Driving with knowl-edge from the physical world. In Proceedings of KDD’ 11,pages 316–325, San Diego, California, USA, 2011.

[224] W. Zhang, K. M. Hansen, and T. Kunz. Enhancing intelli-gence and dependability of a product line enabled pervasivemiddleware. Pervasive and Mobile Computing, 6(2):198–217, 2010.

[225] Y. Zheng, L. Zhang, X. Xie, and W.-Y. Ma. Mining interest-ing locations and travel sequences from GPS trajectories. InProceedings of the international World Wide Web conference(WWW 2009), pages 791–801, Madrid, Spain, 2009.

Documents

Semantic Web 1 (2012) 1–5 IOS Press Semantically-enriched … · 2012-10-30 · Pervasive computing has matured from its origins as an academic research topic to a commercial reality