ABSTRACT

With the exponential growth of textual documents available in unstructured form on the Internet, feature selection approaches are increasingly significant for the preprocessing of textual documents for automatic text categorization. Feature selection, which focuses on identifying relevant and informative features, can help reduce the computational cost of processing voluminous amounts of data as well as increase the effectiveness for the subsequent text categorization tasks. In this paper, we propose a new evident theoretic feature selection approach for text categorization based on transferable belief model (TBM). An evaluation on the performance of the proposed evident theoretic feature selection approach on benchmark dataset is also presented. We empirically show the effectiveness of our approach in outperforming the traditional feature selection methods using two standard benchmark datasets.