Supervised Learning of an Ontology Alignment Process

Abstract

Ontology alignment is a crucial task to enable interoperability among different agents. However, the complexity of the alignment task especially for large ontologies requires automated support for the creation of alignment methods. When looking at current ontology alignment methods one can see that they are either not optimized for given ontologies or their optimization by machine learning means is mostly restricted to the extensional definition of ontologies. With APFEL (Alignment Process Feature Estimation and Learning) we present a machine learning approach that explores the user validation of initial alignments for optimizing alignment methods. The methods are based on extensional and intensional ontology definitions. Core to APFEL is the idea of a generic alignment process, the steps of which may be represented explicitly. APFEL then generates new hypotheses for what might be useful features and similarity assessments and weights them by machine learning approaches. APFEL compares favorably in our experiments to competing approaches.