Information Flow based Ontology Mapping - 2002
- 1. Yannis Kalfoglou IAM Dept. of Electronics and Computer Science University of Southampton Ontology Mapping Marco Schorlemmer CISA Division of Informatics University of Edinburgh Information Flow based ontology mapping
- 2. The Big Picture Local ontology Reference ontology Ontology mappings A theory and a method for Ontology mapping Information Flow based Ontology Mapping
- 3. the need for mapping ontologies: ontology mapping is imminent and necessary; hence, a plethora of related work; but still an ad-hoc process. a theory for ontology mapping: based on a mathematical theory of information flow; applies channel-theory techniques: classifications and infomorphisms. a method for applying IF-Map: includes customised translators; project ontology mappings in XMLified RDF; a Test case: AKT Reference to Soton/Edin Local ontologies. extensions: evolution of ontology maps over time and across organisations; IF-based merging. Overview
- 4. Ontologies nowadays are: Originating from disparate systems often modelling the same domain; Distributed across organisational boundaries; Developed in a variety of knowledge representation formalisms. The advent of the Semantic Web boosted ontologies development: seen as its semantic backbone; That resulted in an ever growing number of ontologies which need to communicate meaning; Which makes mapping imminent to ensure that a concept described in a different manner by two different ontologies has the same semantics; Incidentally, early ontology use suggested that they could be used as as a means to improve interoperability of disparate systems; Today, we have to worry for technologies to improve interoperability of ontologies themselves! Ontology mapping is needed…
- 5. and the community has responded… These challenges aren’t new: in the past 5-6 years many researchers have contributed: Tools: embedded in ontology editors (PROMPT/SMART, Chimeara); Methods: ONIONS, FCA-Merge; Translation systems (OntoMorph), portals (OntoMap), frameworks (MAFRA); Dedicated mechanisms embedded in projects (PhysSys, SKC); Algorithms and heuristics (OntoMediation, PROMPT/SMART, FCA-Merge); Machine learning techniques used in Web systems (CAIMAN, GLUE); Research literature surveys (Visser et.al., Pinto et.al, Uschold et.al); Schema integration techniques borrowed from database community; Links with the Formal Concepts Analysis work (concept lattices, intensions and extensions of concepts) and database schema integration (Schmitt & Saake).
- 6. but ontology mapping is still ad hoc… We observe that most approaches are ad hoc (systems built for another purpose, mapping as a side-effect); translation not mapping tools; use syntactic clues and heuristics (but no semantics); lack a theoretical background; embedded in ontology editors; attached to a specific formalism (often ignore translation caveat); manual and labour intensive (need continuous user feedback – oracle). The situation could be tackled with: A theory of what constitutes ontology mapping; A logic representation of ontology mapping principles; A systematic approach in ontology mapping which should be: Tool and language independent by providing translators to/from imported formats; Easily implemented and deployed on the Semantic Web; Being fully automatic.
- 7. A theory for ontology mapping… Based on channel theory - a mathematical theory of (semantic) information local logics – regularities of components of distributed systems infomorphisms – connections between components: information flow Ontologies are modelled as local logics , represented as classifications: ├─ thing thing ├─ building,car building ├─ thing vehicle ├─ thing car ├─ vehicle building,vehicle ├─ building vehicle car thing thing building vehicle car a 1 1 0 0 b 1 0 1 0 c 1 0 1 1 ontology local logic classification
- 8. which uses classifications and yields infomorphisms Maps of ontologies… thing building vehicle car a 1 1 0 0 b 1 0 1 0 c 1 0 1 1 … are modelled as infomorphisms ,… … and represented as classifications : entity house cottage automobile x 1 0 0 1 y 1 1 0 0 z 1 1 1 0 1 0 1 1 1 1 0 0 1 1 0 0 thing building vehicle car entity automobile cottage house a b c x y z
- 9. How to apply the theory… Number of infomorphisms grows exponentially with the number of concepts; Need to kick-start and constrain the mapping process: Option A: look for matching relation names between local and reference ontologies; traversal of concept hierarchies in order to match their argument types; fix partial map of concept/relation names. Option B: select representative instances of local concepts; classify these instances to reference concepts; fix partial map of instances. Need for: fragmenting reference and local ontologies; monotonic, incremental generation of maps of ontologies.
- 10. … a methodology We have built a stepwise process which consists of: Ontology harvesting (acquire ontologies); Translation; IF-Map (generating infomorphisms); Project mapping (provide XMLified RDF output) Ontology acquisition includes a variety of technologies ranging from web harvesters to ontology libraries and editors; Translation is customised for the purposes of IF-Map method. Currently, we translate to horn logic clauses from RDF, Protégé, Ontolingua, KIF; IF-Map implements the infomorphisms generation; Finally, we project mappings in XMLified RDF format which can be accessed on the Web.
- 12. Test case: AKT Ref/Soton/Edin ontologies We applied IF-Map to map the AKT Refererence ( ref ) ontology to Southampton’s and Edinburgh’s ontologies ( soton and edin ): Ref is encoded in OCML and Ontolingua, soton is in edited in Protégé and edin in Prolog (among other); We used our translators to convert them in Prolog; Soton and edin are populated with instances, ref is not; We mapped fragments of ref to soton and vice versa, and fragments of ref to edin and vice versa; We produce XMLified RDF output showing the generated infomorphisms for concepts and relations.
- 13. Example ref to soton infomorphisms: ref concept document is mapped onto soton concept publication ; ref concept appellation is mapped onto soton concept string ; ref relation publishedby is mapped onto soton relation authoredby ; ref relation hasappellation is mapped onto soton relation title Test case: AKT Ref/Soton/Edin ontologies
- 14. So far, we have experimented with the following mapping scenario: Reference mapped to/from local ontology We also consider trying to: Map a Reference ontology (e.g. AKT-Ref) to another Reference ontology (e.g., IEEE SUO); Map local ontologies when there is no such a thing as a Reference ontology. Mapping scenarios
- 15. Extensions: ontology merging IF-Merge: a method for ontology merging/alignment Scenario: local ontologies and reference ontology(ies); local ontologies “conform” to reference ontology: infomorphisms (IF-Map). Ontology merging: explicit computation of the global ontology ( pushout construction); sharing of knowledge via the reference ontology (virtual ontology).
- 16. Extensions: ontology evolution IF-Merge is sensitive to the way local communities classify their instances infomorphisms: map of concepts + map of instances; global ontology: determined by the interconnection of infomorphisms. A service of automated ontology merging/alignment based on IF-Merge could, thus: automatically update the ontology merging process; refine global ontologies in a way that reflect how local communities classify their instances when using the local ontologies; capture the evolution of ontologies as they are deployed in different contexts.
- 17. IF-Map overcomes some of the problems faced by existing add-hoc ontology mapping efforts: It is purely designed for ontology mapping, merging and aligning; It is based on a sound theoretical foundation: channel theory; It is language and tool independent; It uses translators to accommodate popular ontology formats and produces web accessible XMLified RDF; Future work in three areas: Theory: situating and comparing IF based approach with others (classical logic, FCA, etc.); Methodology: exploring heuristics for kick-starting and constraining the IF-Map/Merge method; Implementation: Moving from a prototype to a full-fledge ontology mapping/merging service deployed on the AKT Bus. Conclusions