关键词: 中文信息处理, 特征扩展, 倾向性分析, 构造分类器

Abstract: This paper presents an extension-based two-step text orientation analysis method. This method uses sentiment words including orientation word list, negative word list and adverb of degree list to extend features of the training texts, and then constructs the classifier CF1 and the classifier CF2 according to whether sentiment words and content words are used in the same way or not. At the classification time, extend features of the testing texts in the same way as for the training texts and classify them with the classifier CF1. If the result of classification is reliable, make a judgment；if not, conduct the second classification for the testing texts with the classifier CF2. Experimental results have proved the effectiveness of the method.

Key words: Chinese information processing, features extension, orientation identification, constructing classifier