On Semantic Similarity and Relatedness for Knowledge-Driven Discovery in Biomedical Data

A great variety of tasks, from word sense disambiguation and document retrieval to assessing the functional similarity of gene products and validating protein-protein interaction networks, depend on the ability to measure the semantic similarity between concepts organized in ontologies. This report is a comprehensive study of classic and recent computational methods measuring semantic relatedness. Motivational arguments set the stage for a survey of the methods, applications and a critical assessment of the methods and of the various evaluation strategies adopted in the literature. Proposals for future directions in the areas of measuring semantic similarity and relatedness, as well as suggestions for improvement, curently under investigation, are offered.