Tradeoffs in Measuring Entity Similarity for Pattern Detection in OWL Ontologies

Syntactic regularities are repetitive structures in the asserted axioms of an ontology represented as generalisations, which are axioms with variables. The Regularity Inspector for Ontologies (RIO) is a framework for detecting such regularities in ontologies. Established clustering techniques are applied to the signature of the ontology to detect clusters of similar entities. Clustering depends on pairwise entity distances, which determine the similarity of two entities. In this paper we present three variations on similarity definition that affect pairwise distances and thus the regularities detected. Our analysis explores and compares methods that capture regularities of different granularity; in particular we analyse commonalities and differences between the generalisations and clusters that result from the three variations of similarity and check if they capture dominant patterns in the ontology in the same way. We perform the analysis using the BioPortal corpus and we discuss the tradeoffs of each similarity function.