Topic Modeling for Linked Open Vocabularies

One of the major issues still open in ontology reuse is how to help users to find the appropriate ontologies and terms for a certain application or domain of interest. In order to complement current ontology similarity-based techniques, topic modeling has the potential of allowing comparisons among ontologies not only on the basis of their lexical content, but also considering their latent semantic structure, by making emerge the topics that are implicit in their lexical descriptions. In this paper we propose a novel method that extracts the lexical contexts of a set of ontologies, annotates them with external senses connected to the Linked Open Data cloud, and uses these annotated contexts to train a probabilistic topic model. We evaluate the method both in terms of coherence of the extracted topics and in terms of its performance when clustering topic related ontologies.