Abstract: Text orientation identification is widely used in user product comment, public opinion analysis. Research has shown that current text orientation identification need the assistance from external resources, a method is presented which is based on emotional description item combined with improved the method of mutual-information, by means of syntactic analysis, it extracts some matched patterns which can obtain emotional description item. Using the pattern matching and the improved of mutual-information calculate as the eigenvalue, the sentiment orientation of text can be got by the training of classification model. Through a series of experiments the results indicate that this method is significantly improved.

Key words: text orientation, syntactic analysis, emotional description item, mutual-information

摘要： 文本倾向性识别可以广泛应用于用户产品评论、舆情分析等。针对文本倾向性识别往往需要借助外部资源的问题，提出一种基于情感描述项及改进的互信息计算相结合的方法，通过句法分析提取出若干可以获得文本情感描述项的匹配模式，根据模式匹配及计算情感描述项的互信息作为特征值，训练分类模型得出文本的褒贬性。通过对酒店、手机语料集实验后的结果进行分析，该方法具有良好的效果。

关键词: 文本倾向性, 句法分析, 情感描述项, 互信息