An Improved Information Content Measure for Semantic Similarity Based on Biomedical Ontology

Abstract

Many semantic similarity measures have been proposed to determine how similar one concept is to another within the context of an ontology. Recently, researchers use intrinsic information content to compute semantic similarity only based on ontology structure. And these measures show their promising results. Intuitively, the height of a concept has an effect on the IC of the concept. Concepts with smaller height are usually more specialized and have more semantic information. In the paper, we propose a new IC computation measure combining the hyponyms of a concept, the height of each hyponym and the depth of the concept. The proposed measure is evaluated against human semantic similarity scores and compared with the existing measures using a standard biomedical ontology SNOMED CT as the input ontology. Results obtained for two benchmarks show that our measure makes semantic similarity measures better correlated with human judgments.