Achieving Consensus in Ontology Management

1. Chapter 9:Ontology Management Service-Oriented Computing: Semantics, Processes, Agents– Munindar P. Singh and Michael N. Huhns, Wiley, 2005

2. Highlights of this Chapter • Motivation • Standard Ontologies • Consensus Ontologies Service-Oriented Computing: Semantics, Processes, Agents - Munindar Singh and Michael Huhns

3. Motivation • Ontologies provide • A basis for communication among heterogeneous parties • A way to describe services at a high level • But how do we ensure the parties involved agree upon the ontologies? • Traditionally: manually develop standard ontologies • Emerging approach: determine “correct” ontology via consensus Service-Oriented Computing: Semantics, Processes, Agents - Munindar Singh and Michael Huhns

4. Some Standard Ontologies • IEEE Standard Upper Ontology • Common Logic (language and upper-level ontology) • Process Specification Language • Space and time ontologies • Domain-specific ontologies, such as health care, taxation, shipping, … Service-Oriented Computing: Semantics, Processes, Agents - Munindar Singh and Michael Huhns

5. An Example Upper Ontology Service-Oriented Computing: Semantics, Processes, Agents - Munindar Singh and Michael Huhns

6. OASIS Universal Business Language (UBL) Service-Oriented Computing: Semantics, Processes, Agents - Munindar Singh and Michael Huhns

7. Standardization Pros • Even if imperfect, standards can • Save time and improve effectiveness • Enable specialized tools where appropriate • Improve the reach of a solution over time and space • Suggest directions for improvement Service-Oriented Computing: Semantics, Processes, Agents - Munindar Singh and Michael Huhns

8. Standardization Cons • Standardization of domain-specific ontologies is • Cumbersome: standardization is more a sociopolitical than a technical process • Difficult to maintain: often out of date by the time completed • Often violated for competitive reasons Service-Oriented Computing: Semantics, Processes, Agents - Munindar Singh and Michael Huhns

9. Standardization: Proposed Approach • Use standard languages (XML, RDF, OWL, …) where appropriate • Take high-level concepts from standard models: • Domain experts are not good at KR • Lot of work in the best of cases • Work toward consensus in chosen domain Service-Oriented Computing: Semantics, Processes, Agents - Munindar Singh and Michael Huhns

10. Inducing Common Ontologies • Instead of beginning with a standard, develop consensus to induce common ontologies • Assumptions: • No global ontology • Individual sources have local ontologies • Which are heterogeneous and inconsistent • Motivation: Exploit richness of variety in ontologies • To see where they reinforce each other • To make indirect connections (next page) Service-Oriented Computing: Semantics, Processes, Agents - Munindar Singh and Michael Huhns

11. Truck APC Wheel Tire Possibly equivalent Truck APC APC partOf equivalence Wheel equivalence Wheel Tire Relating Ontologies: No Overlap Safety in Numbers Service-Oriented Computing: Semantics, Processes, Agents - Munindar Singh and Michael Huhns

12. Relating Ontologies • A concept in one ontology can have one of seven mutually exclusive relationships with a concept in another: • Subclass Of • Superclass Of • Part Of • Has Part • Sibling Of • Equivalent To • Other (topic-specific) • Each ontology adds constraints that can help to determine the most likely relationship Service-Oriented Computing: Semantics, Processes, Agents - Munindar Singh and Michael Huhns

13. Initial Experiment:55 Individual Simple Ontologies about Life Service-Oriented Computing: Semantics, Processes, Agents - Munindar Singh and Michael Huhns

14. 55 Merged Ontologies Service-Oriented Computing: Semantics, Processes, Agents - Munindar Singh and Michael Huhns

15. Methodology for Merging and Reinforcement • Merging used smart substring matching and subsumptionFor example, living  livingThingHowever, living X livingRoombecause they have disjoint subclasses • 864 classes with more than 1500 subclass links were merged into 281 classes related by 554 subclass links • Retained the classes and subclass links that appeared in more than 5% of the ontologies • 281 classes were reduced to 38 classes with 71 subclass links • Merged concepts that had the same superclass and subclass links • Result has 36 classes related by 62 subclass links Service-Oriented Computing: Semantics, Processes, Agents - Munindar Singh and Michael Huhns

16. Consensus Ontology for Mutual Understanding Service-Oriented Computing: Semantics, Processes, Agents - Munindar Singh and Michael Huhns

17. Consensus Directions • The above approach considered lexical and syntactic bases for similarity • Other approaches can include • Folksonomies (as in tag clouds) • Richer dictionaries • Richer voting mechanisms • Richer forms of structure within ontologies, not just taxonomic structure • Models of authority as in the WWW Service-Oriented Computing: Semantics, Processes, Agents - Munindar Singh and Michael Huhns

18. Alternative Approaches We may construct large ontologies by • Inducing classes from large numbers of instances using data-mining techniques • Building small specialized ontologies and merging them (Ontolingua) • Top-down construction from first principles (Cyc and IEEE SUO) Service-Oriented Computing: Semantics, Processes, Agents - Munindar Singh and Michael Huhns

19. Aside: Categorizing Information Consensus is driven by practical considerations • Should service providers classify information where it • Belongs in the “correct” scientific sense? • Where users will look for it? • Case in point: If most people think a whale is a kind of fish, then should you put information about whales in the fish or in the mammal category? Service-Oriented Computing: Semantics, Processes, Agents - Munindar Singh and Michael Huhns

20. Chapter 9 Summary • For large-scale systems development, coming to agreement about acceptable ontologies is nontrivial • Standardization helps, but suffers from key limitations • Consensus approaches seek to figure out acceptable ontologies based on available small ontologies • Should always use standards for representation languages Service-Oriented Computing: Semantics, Processes, Agents - Munindar Singh and Michael Huhns