A Cross-Cluster Approach for Measuring Semantic Similarity between Concepts

Abstract

We present a cross-cluster approach for measuring the semantic similarity/distance between two concept nodes in ontology. The proposed approach helps overcome the differences of granularity degrees of clusters in ontology that most ontology-based measures do not concern. The approach is based on 3 features (1) cross-modified path length feature between the concept nodes, (2) a new features: the common specificity feature of two concept nodes in the ontology hierarchy, and (3) the local granularity of the clusters. The experimental evaluations using benchmark human similarity datasets confirm the correctness and the efficiency of the proposed approach, and show that our semantic measure outperforms the existing techniques. The proposed measure gives the highest correlation (0.873) with human ratings compared to the existing measures using the benchmark RG dataset and WordNet2.0