According to the defects and deficiencies of the classical feature selection method in Chinese comment text, an improved method is proposed for Chinese emotion feature selection. The current existing emotion feature selection method generally only used the statistical information of the feature items in the classes, ignoring the influence of the emotional polarity value on feature selection. Meanwhile the negative words in the sentiment text can cause the reversal of the emotional polarity of the characteristics, which would bring great negative effects on the feature selection. To tackle these problems, antisense transformation processing of the emotion characteristic words is performed in the range of the negative words, which effectively solves the emotional polarity reverasl in the sentiment text. The paper also introduces the emotional polarity values and its frequency into the chi square model (CHI) to improve the effect of CHI on the emotion feature selection. The experimental results show that the proposed method can improve the accuracy of emotion classification by about 1.5% in multiply domain data sets, compared with other algorithms.
