ABSTRACT

Semantic similarity refers to the concept by which a set of documents or words within the documents are assigned a weight based on their meaning. The accurate measurement of such similarity plays important roles in Natural language Processing and Information Retrieval tasks such as Query Expansion and Word Sense Disambiguation. Page counts and snippets retrieved by the search engines help to measure the semantic similarity between two words. Different similarity scores are calculated for the queried conjunctive word. Lexical pattern extraction algorithm identifies the patterns from the snippets. Two machine learning approaches- Support Vector Machine and Latent Structural Support Vector Machine are used for measuring semantic similarity between two words by combining the similarity scores from page counts and cluster of patterns retrieved from the snippets. A comparative study is made between the similarity results from both the machines. SVM classifies between synonymous and non-synonymous words using maximum marginal hyper plane. LS-SVM shows a much more accurate result by considering the latent values in the dataset.