Multiple concept learning - A novel approach to feature selection in text categorization

Abstract

Concept learning is an interesting problem in machine learning and has many applications in real-world problems. This paper considers the multiple concept learning which are extended from binary concept learning. Our main contribution in this paper is to propose a new framework for multiple concept learning. To this end, two sparseness and semantic measures are proposed in order to characterize the scatter and the concentration of concepts in a system. Using both of these two measures, a general strategy of multiple concept learning is given and applied to feature selection in text categorization problem. The experimental results implemented to two benchmark datasets 20Newsgroups and Reuters-21578 show that our approach improve the performances using the Rochio and naive Bayes algorithms compared to conventional methods in the system.