Semantic Similarity Calculation Method using Information Contents-based Edge Weighting

Abstract

In this paper, we propose Semantic Similarity calculation measurement using INformation contents on EdGEs of ontology (SSINEGE) which is a hybrid edge- and information contents-based methodology. SSINEGE is devised to solve the limitation of the applying the same weighted edges by edge-based similarity. So, SSINEGE adopts information-contents theory to calculate the varied weights of edges. The varied weighted edges by SSINEGE can also solve a problem with the same degree of similarity for all pairs of concepts that are sharing a same Least Common Subsumer (LCS). To minimize the overlapped information-contents on the weighted, SSINEGE adopts the conceptual path between concepts instead of depths of the ontology. To verify the superiority of SSINEGE, we compared SSINEGE with widely used four similarity measurements including Leacock and Chodorow. We conducted two kinds of evaluations: first is calculation of similarity using the varied edge-weighting and second is for the discriminative capability using conceptual distances between comparative concepts. To verify the superiority of SSINEGE, we compared the calculated similarities of SSINEGE with Leacock and Chodorow. As the results, we verified that the calculated similarity of SSINEGE is significantly increased than the other comparatives.

Keywords

• Hybrid semantic similarity
• Ontology
• Edge-based semantic similarity
• Information Contents-based semantic similarity