Measuring Semantic Similarity between Words Using Web Documents

Abstract

Semantic similarity measures play an important role in the extraction of semantic relations. Semantic similarity measures are widely used in Natural Language Processing (NLP) and Information Retrieval (IR). The work proposed here uses web-based metrics to compute the semantic similarity between words or terms and also compares with the state-of-the-art. For a computer to decide the semantic similarity, it should understand the semantics of the words. Computer being a syntactic machine, it can not understand the semantics. So always an attempt is made to represent the semantics as syntax. There are various methods proposed to find the semantic similarity between words. Some of these methods have used the precompiled databases like Wor dNet, and Brown Corpus. Some are based on Web Search Engine. The approach presented here is altogether different from these methods. It makes use of snippets returned by the Wikipedia or any encyclopedia such as Britannica Encyclopedia. The snippets are preprocessed for stop word removal and stemming. For suffix removal an algorithm by M. F. Porter is referred. Luhn’s Idea is used for extraction of significant words from the preprocessed snippets. Similarity measures proposed here are based on the five different association measures in Information retrieval, namely simple matching, Dice, Jaccard, Overlap, Cosine coefficient. Performance of these methods is evaluated using Miller and Charle’s benchmark dataset. It gives higher correlation value of 0.80 than some of the existing methods