一种互信息文本特征选择算法的改进

Feature selection is an important part of text categorization, in order to improve the effectiveness and performance of classification, this paper proposed a new approach to improve the Mutual Information based on its shortcomings of inappropriate treatment to negative mutual information and its heavy reliance on low_frequency feature.Statistical reliability is also cosidered. By comparing the new algorithm and the traditional algorithm,experiments show that new algorithm has better results in text classification and Statistical reliability is effective both in Mutual Information and Infomartion Gain.