Towards the integration of KOS with the Linked Data Cloud Vincenzo Maltese, Feroz Farazi DISI – University of Trento
The main message Reusing and sharing ontologies: the linked data initiative In reusing and sharing ontologies: 1. It is fundamental to take into account and make explicit their purpose and semantics 2. The difference in the purpose is reflected in the difference in the semantics and in the applications we enable 3. Before integrating them, it is possible and essential to translate them such that they have the same semantics 4. To maximize reuse, publishing an ontology requires appropriate semantics The Hague - UDC seminar Vincenzo Maltese 2 2011
Building a KOS can be extremely costly o Reuse as much as possible ≡ • institution organization Need : discover similar resources and (partially) integrate them • Solution : • Manual approaches (accurate but slow) • Automatic tools (need background knowledge, need manual validation, but fast) o Share as much as possible • Need : incentives, a common framework • Solution : the Linked Data initiative The Hague - UDC seminar Vincenzo Maltese 3 2011
Ontologies are extremely diverse They may differ in scope, purpose, structure, terminology, language, coverage, formality and conceptualization In sharing and reusing them, it is fundamental to take into account and make explicit: o The difference in the purpose (their goal) o The difference in the semantics (their meaning) The Hague - UDC seminar Vincenzo Maltese 4 2011
Real World semantics Domain of interpretation a building I(a) instance-of Eiffel Tower b I(b) Each term at nodes denotes a real world object or a set of objects The Hague - UDC seminar Vincenzo Maltese 5 2011
Classification semantics Domain of interpretation a building I(a) NT/BT Eiffel Tower b I(b) Each term at nodes denotes a set of documents about real world objects The Hague - UDC seminar Vincenzo Maltese 6 2011
Difference in the purpose is reflected in different semantics
Descriptive ontologies country a d organization is-a is-a instance-of Italy b university e research center f part-of instance-of instance-of University g h Trento c FBK of Trento collaborates-with part-of o Purpose : describing a domain o Semantics : real world semantics The Hague - UDC seminar Vincenzo Maltese 8 2011
Classification ontologies country a d organization NT/BT NT/BT NT/BT Italy b university e research center f d1 NT/BT NT/BT NT/BT University g h Trento c FBK of Trento NT/BT RT d2 d3 d4 o Purpose : classifying, searching and browsing documents o Semantics : classification semantics The Hague - UDC seminar Vincenzo Maltese 9 2011
Difference in the semantics affects what we can do with them
Descriptive ontologies: typical queries country a d organization is-a is-a instance-of Italy b university e research center f part-of instance-of instance-of University g h Trento c FBK of Trento collaborates-with part-of o Give me all the countries: {Italy} o Give me all the organizations: {University of Trento, FBK} o Give me all the organizations located in Italy: {University of Trento} The Hague - UDC seminar Vincenzo Maltese 11 2011
Classification ontologies: typical queries country a d organization NT/BT NT/BT NT/BT Italy b university e research center f d1 NT/BT NT/BT NT/BT University g h Trento c FBK of Trento NT/BT RT d2 d3 d4 o Give me all documents about countries o Give me all documents about Italy: {d1} or {d1, d2} or{d1, d2, d3}? It depends! Do we expand? Are all NT/BT transitive? The Hague - UDC seminar Vincenzo Maltese 12 2011
Making explicit the semantics allows for automation
Descriptive ontologies: how to make them formal country a d organization is-a is-a instance-of Italy b university e research center f part-of instance-of instance-of University g h Trento c FBK of Trento collaborates-with part-of Assume we use Description Logics (DL): o Classes correspond to concepts o Instances correspond to individuals o Is-a relations are translated into subsumption ( ⊑ ) o Other relations correspond to DL roles The Hague - UDC seminar Vincenzo Maltese 14 2011
Classification ontologies: how to make them formal country a d organization NT/BT NT/BT NT/BT Italy b university e research center f d1 NT/BT NT/BT NT/BT University g h Trento c FBK of Trento NT/BT RT d2 d3 d4 Assume we use Description Logics (DL): o Classes correspond to concepts o Documents correspond to individuals o Transitive NT/BT relations are translated into subsumption ( ⊑ ) o RT and non-transitive NT/BT relations correspond to DL roles The Hague - UDC seminar Vincenzo Maltese 15 2011
Converting, integrating and reusing ontologies
Convert before integrating ontologies o It is clearly not appropriate to integrate ontologies having different semantics. o Given the purpose select the semantics • If the purpose is to classify, convert both ontologies into classification ontologies • If the purpose is to describe, convert both ontologies into descriptive ontologies The Hague - UDC seminar Vincenzo Maltese 17 2011
From descriptive to classification ontologies o Convert instances to classes country a o Convert instance-of, is-a and transitive part-of into NT/BT relations instance-of o Convert other relations into RT relations Italy b Hierarchies are constructed on the basis of genus-species (is-a, instance-of) and whole-part (part-of) relations [Ranganathan, 1967. Prolegomena to library classification] country a The process above can be easily automated NT/BT Italy b There is a clear loss of information The Hague - UDC seminar Vincenzo Maltese 18 2011
From classification to descriptive ontologies o Each class has to be mapped to either a real country a world class or instance o Each transitive NT/BT relation has to be NT/BT converted into either an instance-of, is-a or Italy b part-of o Each RT relation and non-transitive NT/BT relation has to be codified into an appropriate real world associative relation country a The process above CANNOT be automated (which relation???) Italy b A substantial amount of human effort is required (instance or class???) The Hague - UDC seminar Vincenzo Maltese 19 2011
For those reasons Distributing schemes as descriptive ontologies would ensure maximum reusability Let us look at a concrete use case…
GeoWordNet: a multilingual descriptive ontology o Neat separation between language and conceptual levels o It is currently in English and Italian o Built from WordNet, Italian MultiWordNet and GeoNames Objects Quantity Classes 110,459 Instances 6,927,078 Instance-of 6,927,078 Is-a 89,266 transitive part-of 5,325 Associative relations 98,907 o We compute the transitive closure for both the descriptive and classification version of the ontology o We use the appropriate semantics according to the task The Hague - UDC seminar Vincenzo Maltese 21 2011
Conclusions o There is the need to reuse/share ontologies o It is fundamental to take into account and make explicit their purpose and semantics o Particular attention has to be paid to the transitivity of the relations o Storing ontologies in their descriptive version maximizes reuse and effectiveness The Hague - UDC seminar Vincenzo Maltese 22 2011
Towards the integration of KOS with the Linked Data Cloud Thank you for your time and interest! Questions? maltese@disi.unitn.it GeoWordNet open source: http://geowordnet.semanticmatching.org/ The Hague - UDC seminar Vincenzo Maltese 23 2011
Extra slides The Hague - UDC seminar Vincenzo Maltese 24 2011
Descriptive ontologies: how to make them formal country a d organization is-a is-a instance-of Italy b university e research center f part-of instance-of instance-of University g h Trento c FBK of Trento collaborates-with part-of university ⊑ organization university(UniversityOfTrento) collaborates(UniversityOfTrento, FBK) … The Hague - UDC seminar Vincenzo Maltese 25 2011
Classification ontologies: how to make them formal country a d organization NT/BT NT/BT NT/BT Italy b university e research center f d1 NT/BT NT/BT NT/BT University g h Trento c FBK of Trento NT/BT RT d2 d3 d4 university ⊑ organization Trento ⊑ Italy Italy(d1) … The Hague - UDC seminar Vincenzo Maltese 26 2011
Limitations of RDF and SKOS RDF SKOS No support for disjointness No support for disjointness Classes can be treated as instances No distinction between classes and instances (we cannot represent documents) Transitivity of relations cannot be We can define non-transitive NT/BT enforced at the level of entities organization a c building Instance-of Instance-of University of Trento b d University of Trento Is b ≡ d ? The Hague - UDC seminar Vincenzo Maltese 27 2011
Recommend
More recommend
