关键词: 文本情感分析, 特征加权, 文档频率, 情感分类

Abstract: There are two universal factors in term weighting for sentiment analysis：Importance of a Term in a Document（ITD）and Importance of a Term for expressing Sentiment（ITS）. An improved ITS algorithm is proposed by combining two state-of-the-art supervised term weighting schemes which have high classification accuracy. The improved algorithm takes both document frequency（the number of documents in which a term occurs）of specific feature and its proportion in the whole document frequency into account. Thus, features which occur predominantly in many documents of one class can be given relatively higher ITS weights. Experiment results show that the proposed algorithm can improve the performance of sentiment classification.

Key words: sentiment analysis, term weighting, document frequency, sentiment classification