COSS: Cross Ontology Semantic Similarity measure — An information content based approach

Computation of Semantic similarity between concepts play a key role in Ontology mapping, Psycholinguistics, Information Integration and Information Retrieval. A COSS (Cross Ontology Semantic Similarity) measure which follows information content approach and is based on Amos Tversky psychological contrast model for finding the semantic closeness of concepts belonging to different biomedical ontologies. This computational approach exploits knowledge sources such as ontologies, thesauri to quantify the information content (informativeness) which helps to assess the amount of information shared by the compared concepts. The proposed approach is corpus independent and it correlates well with the human judgements. The proposed approach has been experimented with two biomedical ontologies: SNOMED-CT (Systemized nomenclature of medical clinical terms) and Mesh (Medical subject headings) within UMLS Framework and the results are reported. This paper also proposed RRCOSS (Refined Resnik Cross Ontology Semantic Similarity) and RLCOSS (Refined Lin Cross Ontology Semantic Similarity) measures. The proposed three approaches outperform the other computational methods as it achieves the highest correlation of 0.920.