An effective measure of semantic similarity

Abstract

Measuring semantic similarity between two concepts is an important problem in web mining, targeted advertisement and domains that need semantic content matching. Nevertheless, developing a computational method capable of generating satisfactory results close to what humans would perceive is still a difficult task somewhat owed to the subjective nature of similarity. This paper presents an effective measure of semantic similarity between two concepts. It relies on hierarchical structure of WordNet 3.0, and considers not only semantic distance but also depth sum and depth difference between two concepts. The correlation value of the proposed semantic similarity measure compared with the human ratings reported by Miller and Charles for the dataset of 28 pairs of noun is higher than some other reported semantic similarity measures for the same dataset. © 2010 Springer-Verlag Berlin Heidelberg.