Abstract: On the basis of vector space model of text expression，a feature weight computing method for text commendatory-
derogatory orientation discrimination is proposed based on Probabilistic Latent Semantic Analysis（PLSA）.In addition to the word frequency of a feature，the information of its thesaurus and homoionym latently obtained by PLSA is taken in consideration to weight computing.Using the feature selection method based on Fisher criterion，and constructing a classifier with Support Vector Machine（SVM），an experiment is conducted under a Chinese review text corpus with size of 2 739 documents（COAE2008）.The experimental results indicate that the presented weight computing method based on PLSA is effective.

Key words: text commendatory-derogatory orientation discrimination, Probabilistic Latent Semantic Analysis（PLSA）, Fisher discrimination criterion, Support Vector Machine（SVM）

摘要： 在文本的向量空间表示模型下，针对文本褒贬倾向判别问题，提出了一种基于潜在语义分析的特征权重计算方法。除词频信息外，该方法考虑了潜在语义分析所提供的同义词、近义词信息对特征权重的影响。采用基于Fisher判别准则的特征选择方法，以支持向量机作为分类器，在2 739篇语料（2008年中文倾向性分析评测）上进行了实验。实验结果表明，提出的特征权重计算方法对文本褒贬倾向判别是有效的。

关键词: 文本褒贬倾向判别, 概率潜在语义分析, Fisher判别准则, 支持向量机