WEST: Weighted-Edge Based Similarity Measurement Tools for Word Semantics

Our recent study discovers that humans are more sensitive to the semantic difference caused by categorization than specification. Based on this new discovery, this paper proposes a novel weighted edge approach embedding the specification levels of both words and their Least Common Ancestor (LCA) into a weighted graph distance by exponentially decreasing the weight along its specification level. Experimental studies show that the similarity value obtained by this new method closely matches human perception. Based on this new method, online tools for word similarity measure are implemented as Web services.