A Compliance Evaluation of a Hybridized Semantic Similarity Measure for Ontology Matching

Semantic similarity measures are used to compute the common shared information between two concepts in domain resources such as ontologies. These measures form the basis of all ontology matching algorithms because they measure the degree of semantic relatedness between matched ontologies. There are two basic categories of semantic similarity measures used in ontology matching process. These measures include the structure based measure and the information based measure. The structure based measures deal with the structure of the ontology, that is, the shortest path length between two concept nodes in the ontology while the information based measures use the Information Content (IC) of concept nodes in the ontology. Despite the importance of semantic similarity measures in ontology matching, their accuracy still remains a challenge.

Aims: Consequently, this paper proposes a hybridized semantic similarity measure which combines the features of both structure and information based measures. The paper also evaluates the proposed measure against existing measures using compliance measures.

Methodology: The proposed measure was implemented using a subset of the Gene Ontology (GO) as the domain specific ontology. This was with the aim of measuring the semantic similarities between muscular development terms and its hierarchies in the GO. The proposed measure was compared with Lin and Wu and Palmers’ measures and evaluated on twenty one (21) pairs of the GO terms using precision, recall rate and F-measure.

Results: The result of the compliance evaluation showed that the proposed measure had a precision rate of 0.86, a recall rate of 1.00 and an F-measure of 0.92 as opposed to the Lin measure and the Wu and Palmer semantic similarity measures which had precisions of 0.83 and 0.79, recall rate of 0.53 and 0.61 as well as F-measures of 0.56 and 0.69 respectively.

Conclusion: The result of the compliance evaluation showed that the proposed measure performed better than Lin measure and Wu and Palmers’ semantic similarity measure. However, the proposed measure needs to be evaluated using a larger data set in order to validate its reliability.