A document classification framework for efficient retrieval

Abstract

Document classification has become an evolving field of exploration with the significant rise in the volume of computerized information. Weighting of a term is an elementary research issue in document classification. Several alternatives to the traditional techniques to weight a term like TF_IDF have been proposed by the researchers. This paper introduces a novel method to weight a term by calculating the semantic similarity between the category label and the term. Also the proposed term weighting technique includes the co-occurrence relation between the terms. Experiments were carried on the 20 Newsgroups and Reuters_21578 benchmark datasets. The results obtained infer that the proposed method outperforms the other weighting methods using various classifiers.