A Hybrid Method of Feature Selection for Chinese Text Sentiment Classification

Text sentiment classification can be extensively applied to information retrieval, text filtering, online tracking evaluation, the diagnoses of public opinions and chat systems. In this paper, a kinds of hybrid methods, based on category distinguishing ability of words and information gain, is adopted to feature selection. For examining the impact of varying the feature dimension to classification results, using corpus of car reviews, feature dimensions, 1000, 2000 and 3000 are adopted in our experiments. The experiments classification results indicate that the hybrid methods are best with feature dimension equal to 3000, and the result by using hybrid methods is superior to that by directly using information gain. In our experiments F value can achieve over 80%. Finally, some mistake examples are employed to indicate the limitations of methods in this paper.