Using Personal Traces in Context Space:

Towards Context Trace Technology

Odd-Wiking Rahlff1, Rolf Kenneth Rolfsen1 and Jo Herstad2

1SINTEF Telecom and Informatics, Blindern, Oslo, Norway; 2University of Oslo, Oslo, Norway

Abstract: Wearables are often described with a focus on providing the user with wearable information access and communication means.The contextual information retrieval aspect is, however, an essential feature of such systems, as in, for example, the Remembrance Agent [1]where manually entered search-terms are used for presenting relevant situational information, or as in different location-based systems [2].In this position paper we outline a general framework of contextually aware wearable systems, and suggest how such mechanisms,collecting massive traces of the user context, may lead to several other interesting uses in what we will call context trace technology.

Keywords: Context matching; Context space; Context trace technology; Contextual awareness; Wearable computer

I keep six honest serving menThey taught me all I knew:Their names are What and Why and WhenAnd How and Where and Who.

[Rudyard Kipling, Just So Stories, 1902]

1. Personal Context

Miniaturisation of personal communication ap-pliances makes it possible to construct wearablecomputers characterised by being (in principle)always on, being context sensitive, etc. asdescribed in the wearable computer FAQ [3].

An important aspect of wearables is that ofthe personal context. This can be defined briefly asa snapshot of the state of the most importantsituational parameters: personal identification,time, location, task at hand, nearby objects,nearby people, etc.

Figure 1 is an example record where thesedata could be gathered as seen from a person-centred point. For mobile users we focus on andrecord the user context, but physical objects andplaces could ultimately also be empowered withcontext recording abilities.

As can be seen, some of the fields are essential(person, time, location), while others are ratherarbitrary (e.g. temperature). Some fields may beupdated automatically, e.g. time, temperature;others can be set semi-automatically, like loca-tion, while some will need manual input, e.g. text

input (which actually could be the user’s inter-action stream with other devices). The morefields that can be derived automatically, thebetter; e.g. by filling in relevant text input bymechanisms such as automatic speech recogni-tion. Also the context should be extensible,when for instance a user adds a biometric sensorand wants this context recorded as well.

Some of the fields of the context record maybe derived by looking up other people’s records,if they grant you access to these or parts of them.An example of this is ‘‘people nearby’’ that canbe derived by checking out whether other peoplewere present at the same place at the given time.

50

# Springer-Verlag London LtdPersonal and Ubiquitous Computing (2001) 5:50–53

Fig. 1. Person-centred context record.

The significant privacy problems and challengesthat arise from such kind of mechanisms will notbe addressed here.

2. Context Matching

In a given context, we intuitively see whichother contexts may be within reach from thatcontext. We may, for example, enter a meeting,and our perception of the hushed silence thereimmediately affords the context of our leaving orsilently joining. We believe that this contextmatching ability is very essential and may beenhanced by personal information appliances.

In Fig. 2 the user is situated in a context c1,which they associate with the actual or possiblecontexts c0, c2, and c42.

A long-term goal for the wearable communitycould be to create a companion system thatsomehow replicates this kind of process, usingartificial sensors and contextual displays, so theuser may experience an enhanced and, ideally,non-obtrusive perception of the world. (Thesystem may, of course, also empower the userwith additional abilities, such as remote seeingand talking, i.e. video telephony and telephonecommunication, etc.) The user of such technol-ogy would ideally perceive the system as being anextension of the awareness functionality of hisown brain.

3. Traces in Context Space

In order to match the user’s current context withother contexts, these contexts must be repre-sented like trigger rules as described by Brown [4]or, as we suggest, have been recorded in a formatsuitable for matching. Our idea differs fromtrigger rules in that the system uses the contexttraces of several users simultaneously. We maythink of this recording as a continuous logging of

a trace in the multidimensional context space(Fig. 3).

Each wearable should have a mechanism, asilent logger, that unobtrusively logs contextrecords as pearls on a string, and stores themsafely for later use [5]. Context logging is alsoidentified as an important architecture require-ment for context-aware systems by Dey et al [6].A context trace could be collected over years,comprising a total record of a lifetime of usage,and collected by many users simultaneously intoa collective database. Such a database would bemore useful the more users populate it using trulywearable equipment that is able to do the silentlogging unobtrusively.

The context space is a generalisation of thetime-geography ideas of Hagerstrand from 1970[7] but now with a possible technological basisfor implementing the data collection and withan awareness purpose. It also may be viewed as avery downscaled precursor of the visionary ideasin the Soul Catcher project from British Tele-com [8].

3.1. The context timeline

Ordering the logged context chronologicallygives rise to a context timeline. If the silentlogging is integrated with a time planner, thelogging can be seen as the convergence pointwhere possible future contexts (e.g. attendingtwo simultaneous interesting events) are collap-sing into the recorded past as the personalcontext trace is spun, as shown in Fig. 4.

Note that we may now derive differentcontext timelines based upon the personalcontext timelines: a context timeline for a location;Louvre being populated by all the visitors there,etc. Another is a context timeline for a given groupof people. Or one may wish to view context spacethrough a view as a map, thereby projecting the

51

Fig. 2. Using our brain vs. a wearable for context matching.

Fig. 3. Context trace of four context records in a (3-dimensional) context space.

Using Personal Traces in Context Space: Towards Context Trace TechnologyUsing Personal Traces in Context Space: Towards Context Trace Technology

context traces onto the location axis, showingthe contexts spatially.

4. Personal Context Trace andContext Matching

Personal requests like those shown in Table 1(left column) can be translated into querieswhere, for example, for the first query thecontext field ‘‘time’’ is left blank for the queryto fill. By keeping some fields constant whilereleasing other fields, similar context records canbe found.

If access to other people’s traces is granted,the traces in context space could be used for thetypes of queries like those shown in Table 1(right column).

Being able to answer such questions willenhance the user’s context awareness [9]. We willrefer to context-aware technology that usesmatchings between massively recorded dynamiccontextual traces, and notices the user ofcontexts nearby as context trace technology(CTT). In a way CTT constitutes a social andsituational information filtering parallel to staticsearch motors on the web, and may one day

prove as useful in a social context provided thecritical mass of users is reached.

It can be noted that classic location-basedinformation systems belong to a subclass of CTTsystems.

The contexts that are closer to your owncurrent context (c4 in Fig. 5) according to someclustering metric where each context field hassome predefined ‘‘closeness-delta’’ are assumed tobe most relevant for the user (cn and cm in Fig.5). These can be the user’s own earlier context,or the contexts of people being simultaneously atthe same location, for example. Also, it wouldprobably be a good idea to give some contextfields more importance/weight than others. Forinstance, the context field ‘‘location’’ is certainlyalmost always more relevant than the contextfield ‘‘temperature’’.

Having the CTT system present the closestcontext may lead to serendipitous discoverywhere, for example, you suddenly detect thecontextual ‘‘presence’’ of somebody else workingsimultaneously on the same task as you, andgoing to the same meeting tomorrow.

Some of these queries, like those involvingsocial navigation, are much more computationintensive, as they are based on statistical datafrom context traces.

5. Communicating in Context

Teleconversations are hampered by the mutuallack of contextual knowledge of those commu-nicating [10]. Therefore, the communicationshould preferably contain a pointer to therespective current context in order to improvecommunication.

52

Fig. 4. Possible future contexts ‘‘condensing’’ into therecorded past context trace.

Table 1: Examples of personal queries and context matching

Personal Context Trace Context Matching

. When was I here last?

. What tasks did I do whenOdin was last around?

. What tasks do I usually dowhen Odin is around?

. When did I last work withtext input ‘‘wearables’’?

. Where did I go from thisplace last time I was here?

. Where do I usually havefun?

. Who was in Hague withme?

. Which objects do I interactwith at The Louvre?

. Who else is writing about‘‘wearables’’ right now?

. What do people usually dohere?

. Where are the others?

. Who should have beenhere, when we do this task?

. When do people usually goto lunch here?

. Where are the nearestpeople interested in ‘‘art’’?

. Which places do peopleinterested in ‘‘art’’ gather?

. What did Liv, Odin and Ido together last year?

Fig. 5. Sphere of contextual awareness (only 3 dimensions forillustration purposes).

O.-W. Rahlff et al.

5.1. Research challenges in CTT

The concepts mentioned spin several researchand design challenges such as:

. What are the minimal core fields to include inthe context record?

. What relevant fields may be included in acontext record?

. How can the context record be extended withuser-defined fields on the fly?

. What defines the efficient granularity of thecontext space?

. How should a global accessible context spacedatabase be implemented?

. What standards should be proposed for traces?(XML?)

. What are the natural input/output views ofcontext queries and replies (maps, timelines,iconic displays, etc.)?

. How do you visualise context space and yourenvironment?

. How should the personal traces be keptpersonal or shared between groups? Whoshould ‘‘own’’ the traces?

. How should queries be mapped to contextattributes, and what happens when a contextfield is released in a query?

. What efficient cluster analysis and metricshould be used for enabling this technology?

In SINTEF we now work on an internal projectin mobile informatics, LAMA, which will beaddressing some of these issues through experi-mental demonstrators, as well as providing amore formal representation of CTT systems.

6. Conclusion

The CTT concept of logging the context trace ofthe user of a wearable and using mechanisms ofcontextual matching in this space seem fertile.

We believe that such systems will ultimatelyaugment our lives profoundly. It is our hope thatthe suggested framework will contribute to morefocused research in this interesting area.

References

1. Rhodes B, Starner T. Remembrance agent: a continu-ously running automated information retrieval system.The Proceedings of The First International Conferenceon The Practical Application of Intelligent Agents andMulti Agent Technology (PAAM ’96), London, UK,April 1996, pp. 487–495. MIT. Available from: http://rhodes.www.media.mit.edu/people/rhodes/Papers/remem-brance.html

2. Rahlff OW, Rolfsen RK and Stegavik H. Where am Inow, computer? Presentation at i3 Annual Conference,Nyborg, Denmark, June 1998. Available from: http://www.informatics.sintef.no/~owr/Publications/i3AC98.ppt

3. MIT Wearable computing FAQ, 1997. Availablefrom: http://www.media.mit.edu/projects/wearables/FAQ/FAQ.txt

4. Brown PJ. Triggering information by context. PersonalTechnologies 1998; 2: 1–9

5. Rahlff OW, Rolfsen RK, Herstad J. The role of wearablesin social navigation. In: Munro AJ, Hook K, Benyon D(eds) Social navigation of information space. CSCWSeries, Springer-Verlag, London, 1999; 217–236

6. Dey AK, Salber D, Futakawa M and Abowd GD. Anarchitecture to support context-aware applications. GVUTechnical Report, GIT-GVU-99-23.ftp://ftp.cc.gatech.edu/pub/gvu/tr/1999/99-23.pdf

7. Parkes DN, Thrift NJ. Times, spaces and places: achronographic perspective. In: Time-geography: theLund approach. Wiley & Sons, Chichester, 1980:650

8. Cochrane P. British Telecom. Soul catcher. 1997.Available from: http://www.labs.bt.com/library/cochrane/papers/SoulCatcher.htm

9. Rolfsen RK, Jørgensen HD, Carlsen S. Contextualawareness: survey and proposed research agenda.SINTEF Telecom and Informatics, 1999; http://www.informatics.sintef.no/projects/awareness/ECSCWAware-ness.doc

10. Rahlff OW, Rolfsen RK, Herstad J, Thanh DV Contextand expectations in teleconversations. In: Proceedings ofHCI International ’99, vol. 2 (8th InternationalConference on Human-Computer Interaction).Munich, Germany, 22–27 August 1999; 523–527

Correspondence to: Odd-Wiking Rahlff, SINTEF Telecom andInformatics, PO Box 124 Blindern, N-0314 Oslo, Norway.Email: owr@informatics.sintef.no

53

Using Personal Traces in Context Space: Towards Context Trace TechnologyUsing Personal Traces in Context Space: Towards Context Trace Technology