Paradox in Applications of Semantic Similarity Models in Information Retrieval

Abstract

Semantic similarity models are a series of mathematical models for computing semantic similarity values among nodes in a semantic net. In this paper we reveal the paradox in the applications of these semantic similarity models in the field of information retrieval, which is that these models rely on a common prerequisite – the words of a user query must correspond to the nodes of a semantic net. In certain situations, this sort of correspondence can not be carried out, which invalidates the further working of these semantic similarity models. By means of two case studies, we analyze these issues. In addition, we discuss some possible solutions in order to address these issues. Conclusion and future works are drawn in the final section.