Abstract: A novel hierarchical text sentiment classification approach based on CCA-VSM classifier and kernel Fisher discriminant is proposed to improve classification accuracy. CCA is utilized to reduce the dimensionality of feature vectors. And then vector space model is built on reduced vector set. By doing this, a novel CCA-VSM classifier is proposed according to the diversity between VSM models. [R] models, which possess smaller diversity, would be selected by CCA-VSM classifier. Kernel Fisher discriminant is used to make judgment. Experiment results show that hierarchical classifier is superior to SVM in text sentiment classification problem, and also show that the method of weight computation and the rule of parts of speech feature selection have big effection on classification results.

Key words: text sentiment classification, kernel Fisher discriminant, Support Vector Machine（SVM）, vector space model, canonical correlation analysis

摘要： 针对文本情感分类准确率不高的问题，提出基于CCA-VSM分类器和KFD的多级文本情感分类方法。采用典型相关性分析对文档的权重特征向量和词性特征向量进行降维，在约简向量集上构建向量空间模型，根据模型之间的差异度设计VSM分类器，筛选出与测试文档差异度较小的[R]个模型作为核Fisher判别的输入，最终判别出文档的情感观点。实验结果表明：该方法比传统支持向量机有较高的分类准确率和较快的分类速度，权重特征和词性特征对分类准确率的影响较大。

关键词: 文本情感分类, 核Fisher判别, 支持向量机, 向量空间模型, 相关性分析