vatant

Slide1: 

Ontology-Driven Topic Maps XML Europe Conference Amsterdam, April 20, 2004 Bernard Vatant bernard.vatant@mondeca.com

Disclaimers: 

Disclaimers This presentation is *not* … Yet another RDF to Topic Maps transformation Although it deals with using together Topic Maps and the RDF dialect OWL Writing ontologies in Topic Maps format Although it claims this would be a not-so-good idea Proposing an alternative to Topic Map Constraint Language Since TMCL is not yet here This presentation is about ... What it might mean for a Topic Map to « commit to an ontology » And singularly to an OWL ontology Hints on how such a commitment can be declared and validated Benefits, drawbacks, pitfalls, limits of such an approach Including use of Topic Maps in the general Semantic Web framework Content is somehow controversial So, I ’ll keep it short to let room for flames and otherwise comments

Ontologies and Topic Maps as KOS: 

Ontologies and Topic Maps as KOS A Knowledge Organisation System organises knowledge objects More or less everything can be considered as a knowledge object An ontology Defines classes and generic properties of knowledge objects Defines constraints on those classes and properties Generally uses formal semantic declaration for those definitions Can be checked for logical consistency A Topic Map Defines a specific set of knowledge objects and their individual properties Can include topics used as classes of the above Has no formal semantics Makes no provision for logical consistency

Where Ontology and Topic Map overlap: 

Ontology Topic Map Where Ontology and Topic Map overlap Constraints Classes Instances Data 1953-04-21 Bernard Mondeca A R1 R2 Organisation Employee Employer Employment Person In an Employment Association The Employer role is played by an Organisation The Employee role is played by a Person homePage birthDate www.mondeca.com

Ontologies and Topic Maps in XML: 

Ontologies and Topic Maps in XML Different languages developed by different standard bodies Ontologies : OWL-RDF is W3C recommendation Topic Maps : XTM is part of ISO 13250 Interoperability elements XML syntax URIs used as objects identifiers Classes, SubClasses, Instances Similar semantics using different expression for Linking Knowledge Objects together OWL ObjectProperty, XTM Association-Role Linking Knowledge Objects to Data OWL DatatypeProperty, XTM occurrence

Classes in OWL-RDF and in XTM: 

OWL-DL Classes and properties must be declared Classes, properties and instances are distinct OWL-Full or RDF No declaration of classes and properties is necessary Any RDF Resource can be used as class or property XTM is similar to OWL-Full No declaration of classes is necessary Any XTM Topic can be used as a class (or type) For topics, associations, roles or occurrences. Any RDF Resource can also be used as subject indicator for a type Local declaration of types is optional Classes in OWL-RDF and in XTM

Local Declaration vs External Reference: 

Ways to declare the class of a topic in XTM Reference to a topic element in the local topic map <topic id="person"> <topic id="dupont"> <instanceOf> <topicRef xlink:href="#person"/> </instanceOf> </topic> Reference to an external subject indicator <topic id="dupont"> <instanceOf> <subjectIndicatorRef xlink:href="http://www.example.org/ontology#Person"/> </instanceOf> </topic> Local Declaration vs External Reference

Levels of Ontological Commitment: 

Case #1: « Topic class » is defined as a topic in the local Topic Map The class indication has no formal consequences The reference class is only a flat identifier Will be used only to support merging of classes Case #2: « Topic class » is defined by a subject indicator Dereferencing this URI should yield at least some subject definition What this definition should consist of is yet to be specified Work on PSI content definition is in standby In the general case, only information for human consumption Case #3: « Topic class » has a formal definition in an ontology The topic instance is expected to « commit » to the definition of the class What this includes exactly has to be further specified Definition of SubClasses, Properties Domain, Range and Restrictions ... Levels of Ontological Commitment

Local vs Global Commitments: 

Local Commitment is tricky to handle Topic X is declared instance of Class A in ontology O1 ... Topic Y is declared instance of Class B in ontology O2 ... What global sense can be made of such declarations? What if O1, O2, On … are independently managed? What if they are mutually inconsistent? Global Commitment is more constraining but more straightworward « A Topic Map T globally commits to the ontology O » iff Any class explicitly used in T is defined in O Including topic classes, association, role and occurrence types Assertions in T are globally consistent with constraints defined in O What this includes exactly has to be further specified ... e.g. « If X is instance of A, then X is instance of all superclasses of A » Or it can be let to applications to decide which constraints they manage Local vs Global Commitments

Relationship Classes: 

Same purpose and use cases in OWL and Topic Maps Link knowledge objects through semantic assertions ex : « Mondeca employs Bernard » Different ways to express it Object Properties have binary definition hasEmployee (Mondeca, Bernard) An association is defined as a set of role members Employment (Employer: Mondeca ; Employee: Bernard) No notion of Domain, Range, Restriction in XTM association roles Anything can be linked to anything by any type of association Role Types used for an Association Type are not constrained Classes of role players are not constrained Relationship Classes

Association Templates in OWL: 

Association Templates in OWL Defining an Association Template in OWL An Association Type is defined as an owl:Class (Employment) A Role Type is defined as an owl:ObjectProperty (Employer, Employee) Association Type is the Domain of the constrained Role Types The Role Type Range constrains the class of role players <owl:Class rdf:about="http://www.example.org/ontology#Employment"> <owl:ObjectProperty rdf:about="http://www.example.org/ontology#employer"> <rdfs:domain rdf:about="http://www.example.org/ontology#Employment"/> <rdfs:range rdf:about="http://www.example.org/ontology#Organization"/> </owl:ObjectProperty> <owl:ObjectProperty rdf:about="http://www.example.org/ontology#employee"> <rdfs:domain rdf:about="http://www.example.org/ontology#Employment"/> <rdfs:range rdf:about="http://www.example.org/ontology#Person"/> </owl:ObjectProperty>

Logical vs Physical Data Types: 

Data Types define at the same time or separately The logical types of data attached to a given type of Knowledge Objects ex : A « Person » may have a « birthDate », « CV », « Welfare Number » The physical type of such data ex : « CV » has to be validated against a specific XML schema The logical/physical distinction is not always clear ... In XTM Only the logical type is defined for an occurrence No provision for definition of physical type of occurrence Except through a subject indicator, with no standard interpretation In RDF-OWL Many simple physical datatypes are defined (XML Schemas) No provision for definition of logical datatypes of arbitrary complex structure Logical vs Physical Data Types

Validation of both Logical and Physical Types: 

Ontology Topic Map Validation of both Logical and Physical Types CVdupont.xml Dupont Person ComplexData Curriculum-Vitae CV_Schema.xsd hasValue cv occurrence schema Logical Type Physical Type Validation

Support and Tools for defining OWL ontologies: 

Support and Tools for defining OWL ontologies OWL is W3C Recommendation since February 2004 http://www.w3.org/TR/2004/REC-owl-ref-20040210/ Protégé Ontology Editor and Community of Practice OWL Plugin available for Protégé-2 http://protege.stanford.edu/plugins/owl/ Semantic Web Best Practices and Deployment Working Group Started in March 2004 http://www.w3.org/2001/sw/BestPractices/

Mondeca Use Cases: 

Mondeca ITM is ‘ ontology-driven ’ since 2003 Concept tested in two on-going European IST Projects Hi-Touch : Distributed Resources for Sustainable Tourism MOSES : Knowledge Grid of University Websites On-Going validation with several customer projects Hachette Multimedia Documentation and Knowledge Base Management PSA Management of Data Flow in car conception Dassault-Systems Competitive Intelligence, with Text Mining Tools controlled by ontology Mondeca Use Cases

Limits and Pitfalls: 

Some OWL features have not been used yet in this framework Are all OWL features relevant to TM? Is it realistic to control a TM by any out-of-the-shelf OWL ontology? Does not seem a so good idea Dealing with several standards together Augments the learning curve for adopters Might trigger concerns among TM addicts (not developed here) Trade-off between constraints and expressivity Keep ontologies light Too complex ontologies are barely usable and not re-usable Too many constraints will kill TM expressivity Limits and Pitfalls

Perspectives: 

Some paths to explore ... In Topic Maps land Could TMCL rely partly on OWL? Or should it? At least for « ontological » constraints? Insertion of Topic Maps in the Semantic Web framework Is generic RDF expression of TM a good idea? Maybe using OWL for TM is a less difficult path? Hint : SW Best Practices and Deployment Group is open to proposals ... Perspectives

Thanks for your attention: 

Thanks for your attention Comments Questions