Constructing a measure of information content for an ontological concept

Abstract

Ontologies have become a focal point in the advancement of the Semantic Web especially in the biological and biomedical domains which have a wealth of ontologies such as those found in BioPortal. Computing the degree of semantic similarity between ontological concepts has been a significant function for their use in various applications. Semantic similarity measures that utilize the information content (IC) of an ontological concept have become more and more standard since they have been widely studied and evaluated. The meaning of information content and its calculation, however, have seen numerous interpretations and formulations. Just recently a method of calculating IC incorporates belief function and plausibility theory into the early corpus-based IC method. The argument is that humans intuitively use inductive inference, and, therefore, plausibility should be incorporated when calculating IC. Various approaches to determine IC measures and the role of the ontology structure has played in IC measures are reviewed. The recent inductive inference approach, which considers both the ontology structure and corpus frequency, is analyzed and compared to other existing IC measures. The analysis and comparison is motivated by the assumptions made in the construction of these IC measures and provides insights into factors to be considered in assessing the IC of an ontological concept.