Semantic Similarity based on Weighted Ontology

Semantic similarity methods have been the focus of interest in linguistics, biomedical and artificial intelligence studies. They have been conceived to compare concepts, facilitate searching through ontologies, and improve matching and aligning ontologies. In the context of Geographic Information Science, similarity methods have been used to measure the degree of semantic interoperability between Geographic Information Systems (GIS) or geographic data. In this paper, we analyze three different methods for evaluating the semantic similarity of geographic concepts (or classes). They are defined on the basis of information content, features' commonality/non-commonality and combined approaches. The first two methods are extensively used in the domain of geographic similarity reasoning and are selected as the most representative methods in our analysis. Accordingly, Lin, Dice, MDSM similarity methods have been recalled and contrasted against our combined GSim method. The GSim method is an ontology-centered and information content-based method that is conceived to capture both the concept similarity within the hierarchy, and the attribute similarity (or tuple similarity) of geographic classes. Thus, the problem of weight assignment to concepts of reference ontology has been considered. To this end, two different approaches, frequency and probabilistic-based methods, have been analyzed. The experimental results illustrate that the GSim method provides more reliable measures for comparing geographic classes with respect to the selected methods.