A blended feature selection method in text classification

In text classification system, accuracy is a major indicator of performance, and feature selection method has a significant impact on it. In this paper, we propose a blended feature selection method, which combines four traditional feature selection methods (document frequency, information gain, mutual information and chi-square) into a better feature selection method. BFSM is tested on a Chinese corpus of 4000 documents and improves the accuracy compared with those traditional methods.