Machine Learning for Ontology Learning

The report presents an investigation of the ontology learning problem. It presents an overview of ontology research field and the discussion of available machine learning techniques and learning problems that arise in ontology learning. Recently appeared Web-based and Business-to-Business ontologies forced new learning problems to appear that are also presented in the report. Introduction The idea of a knowledge base is moving nowadays from small standalone tool heavily maintained by a knowledge engineer to a distributed self-contained piece of knowledge shared worldwide by a community of interested people. Ontologies are developed specially to enable this new type of knowledge representation and recent time they become a popular topic in the Knowledge engineering community. Ontologies Ontologies provide a formal, explicit specification of a shared conceptualization of a domain that can be communicated between people and heterogeneous and widely spread application systems. They have been developed in Artificial Intelligence to facilitate knowledge sharing and reuse [Fensel, 2000]. Ontologies for Knowledge Engineering Ontologies are targeted to solve the same problem as knowledge engineering (KE) and knowledge bases (KB): to store and use the knowledge. What are the differences between ontologies and knowledge bases? General knowledge base is • constructed manually by a knowledge engineer(s); • contains a partial model of a domain; knowledge engineer tries to minimize the KB keeping a big part of domain knowledge in his mind and not encoding explicitly; • the semantics of the concepts used within the KB is not included into the KB being encoded, for example, with the First Order Logic, or LISP; • developed specially for one organization, knowledge sharing and reuse between different organizations is very difficult; • contains the knowledge engineers view on a domain, not necessary a consensus knowledge; • may include descriptions (pieces of knowledge) made in natural language; • not web-based. Unlike knowledge bases ontologies have all knowledge about the domain in one unit (even in one file). The provide: • formal or machine readable representation, no natural language descriptions inside; • full and explicitly described vocabulary of the domain, together with explicit semantics of the concepts; • full model of the domain, including redundant for the task relations between the concepts; • a consensus knowledge, or common understanding of the domain; • developed specially for being shared and reused; • web-based. The paper [Studer et al., 1998] gives an overview about the development of the field of Knowledge Engineering over the last 15 years. The authors discuss the paradigm shift from a transfer view to a modeling view and describe two approaches which considerably shaped research in Knowledge Engineering: Role-limiting Methods and Generic Tasks. To illustrate various concepts and methods that evolved in the last years they describe three modeling frameworks: CommonKADS, MIKE, and PROTEGE-II. This description is supplemented by discussing some important methodological developments in more detail: specification languages for knowledge-based systems, problem-solving methods, and ontologies. The paper emphasis recent shift from traditional KE for KBs into shared and widely reused ontologies. The Languages for Ontology Representation A lot of languages were proposed for ontology representation, see [Fensel, 2000] for an overview. Because language analysis is beyond the scope of this report we will just mention the most popular: • KIF (Knowledge Interchange Format) is a logic-based language; • CYCL project provided a LISP-like protocol for interaction between ontologies; • Frame (logics) provide a wide range of frame-based languages; • OIL (Ontology Interchange Layer) [Fensel et al., 2000] is a frame-based standard proposal for ontology interchange language, that might include the necessary core language primitives and provide the developers with the possibility to create their own extensions of the language. An excellent recent overview of the ontology specification languages is presented in [Corcho&GomezPerez, 2000]. Types of Ontologies In the paper [Studer et al., 1998] the authors present the following four main types of ontologies: • Domain ontologies. They capture valid for a particular type of domain (e.g. electronic, medical, mechanic, digital domain). • Generic ontologies that are valid across several domains. Generic ontologies provide supertheories, like knowledge about is-a or part-of relation. • Application ontologies that contain all the necessary knowledge for modeling a particular domain (usually a combination of domain and method ontologies). • Representation ontologies that provide representational entities without stating what should be represented (ex. Frame Ontology) and do not refer to any particular domain. For example the Frame Ontology defines concepts such as frames, slots and slot constraints allowing expressing knowledge in an object-oriented or frame-based way. Ontologies are very popular in the field of natural language processing (NLP). By now linguistic ontologies form the most successful part of the ontology field. The most widely cited NLP ontology is WordNet [Fellbaum, 1998]: an on-line lexical reference system whose design is inspired by current psycholinguistic theories of human lexical memory. English nouns, verbs, adjectives and adverbs are organized there into synonym sets, each representing one underlying lexical concept. This ontology is available in the Web and provides excellent lexical resource for NLP. Also WorldNet provides a bridge from NLP to different domain ontologies. Another interesting ontology application area is User Interface (Web-site ontologies), see [Perkowitz&Etzioni, 2000] for relevant research and ideas. There are also many other application areas.