Mining Non-Taxonomic Conceptual Relations from Text

Non-taxonomic relations between concepts appear as a major building block in common ontology definitions. In fact, their definition consumes much of the time needed for engineering an ontology. We here describe a new approach for mining nontaxonomic conceptual relations from text building on shallow text processing techniques. We use a generalized association rule algorithm that does not only detect relations between concepts, but also determines the appropriate level of abstraction at which to define relations. This is crucial for an appropriate ontology definition in order that it be succinct and conceptually adequate and, hence, easy to understand, maintain, and extend. We also perform an empirical evaluation of our approach with regard to a manually engineered ontology. For this purpose, we present a new paradigm suited to evaluate the degree to which relations that are learned match relations in a manually engineered ontology.