Aus Aifbportal
Wechseln zu:Navigation, Suche

Supervised Learning of an Ontology Alignment Process

Supervised Learning of an Ontology Alignment Process

Published: 2005 April
Herausgeber: Klaus-Dieter Althoff and Andreas Dengel and Ralph Bergmann and Markus Nick and Thomas Roth-Berghofer
Buchtitel: Professional Knowledge Management: Third Biennial Conference, WM 2005
Ausgabe: 3782
Reihe: LNAI
Seiten: 508 - 517
Verlag: Springer
Erscheinungsort: Kaiserslautern, Germany

Referierte Veröffentlichung


Semantic alignment between ontologies is a necessary precondition to establish interoperability between agents or services using different ontologies. Thus, in recent years different methods for automatic ontology alignment have been proposed to deal with this challenge. Thereby, the proposed methods were constricted to one of two different paradigms: Either, (i), proposals would include a manually predefined automatic method for proposing alignments, which would be used in the actual alignment process. They typically consist of a number of substrategies such as finding similar labels. Or, (ii), proposals would learn an automatic alignment method based on instance representations, e.g. bag-of-word models of documents. Both paradigms suffer from drawbacks. The first paradigm suffers from the problem that it is impossible, even for an expert knowledge engineer, to predict what strategy of aligning entities is most successful for a given pair of ontologies. This is especially the case with increasing complexity of ontology languages or increasing amounts of domain specific conventions. The second paradigm is often hurt by the lack of instances or instance descriptions. Also, knowledge encoded in the intensional descriptions of concepts and relations is only marginally exploited by this way. Hence, there remains the need to automatically combine multiple diverse and complementary alignment strategies of all indicators, i.e. extensional and intensional descriptions, in order to produce comprehensive, effective and efficient semi-automatic alignment methods. Such methods need to be flexible to cope with different strategies for various application scenarios, e.g. by using parameters. We call them “Parameterizable Alignment Methods” (PAM).We have developed a bootstrapping approach for acquiring the parameters that drive such a PAM. We call our approach APFEL for “Alignment Process Feature Estimation and Learning”.

ISBN: 3540304657
VG Wort-Seiten: 16
Download: Media:2005_882_Ehrig_Supervised_Lear_1.pdf
Weitere Informationen unter: Link




Web Science


Maschinelles Lernen, Ontology Learning, Künstliche Intelligenz, Semantic Web