A New Ontology-Based Semantic Similarity Measure for Concepts Subsumed by Multiple Super Concepts

Semantic Similarity relates to computing the similarity between concepts of ontology. There exist four approaches to calculate the semantic similarity. The first approach is based on path length. Under this approach we studied and compared some of the measures on a bench mark dataset. Among the compared measures Wu & Palmer measure has the advantage of being simple to implement and has better performance compared to the other similarity measures. This measure considers only the depth of the LCS: (Least Common Subsumer) we call in our paper as Closest Common Parent for similarity computation. But there are complex and large taxonomies, covering thousands of interrelated concepts including several overlapping hierarchies, and extensive use of multiple inheritances (i.e. a concept is subsumed by several super concepts). For such taxonomies using only the LCS will ignore a great amount of explicit knowledge. To overcome this limitation we propose ontology based semantic similarity measure which extends Wu & Palmer measure by considering ASC :( All Subsumed Concepts). We compared both the measures on two benchmark datasets. The obtained results show that our measure gave improved similarity values compared to the Wu and Palmer measure.