A method for ontology-based semantic relatedness measurement

Abstract

There are many methods having different approaches for assessing similarity and relatedness and they are used in many application areas, including web service discovery, invocation and composition, word sense disambiguation, information retrieval, ontology alignment and merging, document clustering, and short answer grading. These methods can be categorized as path-based, information content-based, feature-based, geometric model-based, and hybrid approaches. These approaches use resources such as concept hierarchy, conceptual graph, and corpus for computing similarity and relatedness. With the rise of the semantic web, ontologies have attracted the attention of several researchers. Ontologies represented in the Web Ontology Language (OWL) are also valuable resources for similarity and relatedness measurement. The method proposed in this paper interprets some OWL constructs to assess semantic relatedness. The motivation behind this is to benefit from the rich expressive power of OWL to obtain better semantic relatedness measurement results. The success of the method has been validated against human judgments. The correlation between human judgments and automatically computed semantic relatedness values was calculated as 0.685 and was significant at the 0.01 level.