15
Ontology Design Pattern Property Specialisation Strategies Karl Hammar Jönköping University / Linköping University

Ontology Design Pattern Property Specialisation Strategies

Embed Size (px)

DESCRIPTION

Presentation held at EKAW 2014. Full paper at http://link.springer.com/chapter/10.1007/978-3-319-13704-9_13

Citation preview

Page 1: Ontology Design Pattern Property Specialisation Strategies

Ontology Design Pattern Property Specialisation

StrategiesKarl Hammar

Jönköping University / Linköping University

Page 2: Ontology Design Pattern Property Specialisation Strategies

ODP Specialisation

• Typically a Content ODP contains/is packaged as a small reusable OWL ontology.

• Using that general/abstract ODP in a specific use case requires specialisation.

• Often done via import and mapping of original OWL ontology.

Page 3: Ontology Design Pattern Property Specialisation Strategies

Research Question

• How are Content Ontology Design Patterns (ODPs) used or specialised in Ontology Engineering projects for the Semantic web, and what are the effects of such usage?

• Motivation: to develop better understanding of ODP usage, supporting development of improved (guided?) tooling for such use.

Page 4: Ontology Design Pattern Property Specialisation Strategies

Method Overview

1. Gather OWL ontologies that reference known ODP namespaces.

2. Extract mapping axioms (e.g. subsumption or equivalency mappings b/w ontology and ODP).

3. Find recurring patterns in mapping axiom structure.

Page 5: Ontology Design Pattern Property Specialisation Strategies

Understanding ODP Specialisation

• 347 non-trivial ontologies gathered through Google Search, LODStats, LOV, OntologyDesignPatterns.org, IKS project

• 41 ODP-using ontologies found, with 107 mapping axioms.

• 85 % of mapping axioms are one-way subsumption of classes or properties. ODP classes/properties are nearly always superclasses/superproperties.

• 20 ODP specialisation modules that specialise object properties.

• Three distinct ways of specialising properties were noted.

Page 6: Ontology Design Pattern Property Specialisation Strategies

Property-Oriented Strategy• Most common way of modeling

property specialisation.

• New subproperties are created.

• Domains and ranges of sub properties narrower than domains and ranges of superproperties.

• Can be partially instantiated (e.g., domain or range are linked to more general term)

ce:Collection owl:Thingce:hasMember

rdfs:domain rdfs:range

cc:Content Collection

cc:Content Item

rdfs:rangerdfs:domain

cc:hasContentMember

rdfs:subClassOf rdfs:subClassOfrdfs:subPropertyOf

Page 7: Ontology Design Pattern Property Specialisation Strategies

ir:InformationRealization

ir:InformationObjectir:realizes

rdfs:domain rdfs:range

wf:Weather Forecast

wf:Weather Information

rdfs:subClassOfrdfs:subClassOf

ir:realizes some wf:Weather Information

owl:equivalentClass

ir:realizes only wf:Weather Information

rdfs:subClassOf

Class-Oriented Strategy• Reuses original property;

constrains the use of that property by class restriction axioms.

• Restricts which (specialised) classes that may be related via a given property, i.e. locally emulates domain/range.

• Not logically equivalent to property-oriented strategy - but can be used to solve similar modeling problem.

Page 8: Ontology Design Pattern Property Specialisation Strategies

Hybrid Strategy

• Does both: new subproperties are defined with domains and ranges and property restrictions on participating classes are also put in place (possibly redundantly).

Page 9: Ontology Design Pattern Property Specialisation Strategies

Specialisation Strategies

Specialisation strategy Occurrences

Property-oriented 9

Class-oriented 6

Hybrid 5

Page 10: Ontology Design Pattern Property Specialisation Strategies

RQs Evolved

• To what degree are the class-oriented, property-oriented, or hybrid property specialisation strategies used in published ontologies?

• What are the reasoning performance effects of specialising in accordance with the class-oriented or property-oriented strategies?

Page 11: Ontology Design Pattern Property Specialisation Strategies

Strategy Use

Specialisation strategy Occurrences Distribution ODP specialisation

distribution

Property-oriented 193 78 % 45 %

Class-oriented 33 13 % 30 %

Hybrid 23 9 % 25 %

Page 12: Ontology Design Pattern Property Specialisation Strategies

Strategy Effects

• Property-Oriented strategy treats properties as first-order citizens that can be annotated and typed.

• Property-Oriented strategy possibly more intuitive for non-expert users?

• Class-Oriented strategy makes resulting ontologies and datasets interoperable to greater degree.

• Class-Oriented strategy causes computationally unfavorable conditions and places ontology outside OWL2 EL.

Page 13: Ontology Design Pattern Property Specialisation Strategies

Strategy Effects

Reasoning task Benchmark Reasoner PO time CO time

Consistency checking BSBM Pellet 1.274 s 1.897 s

Consistency checking BSBM HermIT 1.984 s 27.193 s

Consistency checking LUBM Pellet 8.230 s 42.887 s

Consistency checking LUBM HermIT 10.097 s 46 min

Realising individuals BSBM Pellet 2.389 s 9.482 s

Realising individuals LUBM Pellet 1.801 s 4+ hours

Page 14: Ontology Design Pattern Property Specialisation Strategies

Conclusion• Three strategies for property specialisation are

found both amongst ODPs and ontologies in general.

• Pure Property-Oriented strategy less common among ODP-using ontologies than among ontologies in general.

• There may be tradeoffs between performance and interoperability that ODP users and tooling need be aware of.

Page 15: Ontology Design Pattern Property Specialisation Strategies

INTERESTED IN ONTOLOGY ENGINEERING?

Participate in our survey, contribute to science, and win Amazon gift cards!

http://goo.gl/P1w9gG

$$

Marketing