Applying Graph Partitioning Techniques to Modularize Large Ontologies

Abstract. Nowadays, it is difficult to reuse ontologies, especially those that cover a large domain. It is in this context that ontology modularization can be useful. The goal of this work is to investigate graph partitioning techniques and their application on the modularization of large ontologies, typically, biomedical ontologies. Such investigation may be divided in two steps: (i) how to convert an ontology, represented in OWL or RDF languages into a graph; (ii) which partitioning algorithm would be suitable. More specifically, this work focus is on how to preserve certain ontology properties/relationships in the generated modules. Therefore, a single way of graph conversion was adopted and user-defined edge weights were taken into account. Five graph partitioning algorithms were used for the present investigation, but just three of them were used to verify their behavior in face of edge weight variations. A case study was conducted using a toyontology on the pizza domain, and showed preliminary but interesting results.