中文摘要：

针对用户评论中产品特征-观点对的提取及情感分析问题进行了研究。为了提高提取及分析的准确性，利用组块分析提取产品特征，从中寻找到频繁项集，并用逐点互信息量（PMI）对候选产品特征进行过滤，得到产品的特征集合；利用特征与情感词在位置上的邻近关系，提取情感词并组成特征-观点对，通过点互信息方法（SO—PMI）进行情感倾向分析。为验证该方法的有效性，以酒店评论文本为例，从中提取酒店的特征-观点对并进行情感分析，准确率为76．68％，召回率为70．84％。实验结果表明，引入组块分析可以有效地解决商品评论的细粒度情感分类问题。

英文摘要：

Aiming at the problem of product feature-opinion extraction and sentiment analysis, this paper used Chinese chunk parsing to extract the feature, and generated frequent sets. Then it filtered the candidate product features according to the rules of the minimum support, frequent nouns and PMI. And then it used the adjacency relations in the position of feature and emotion words to extract the opinion words and generated feature-opinion. At last, it used SO-PMI to analyse the sentiment. In order to verify the effectiveness of this method, it used the hotel reviews, the precision rate reached 76.68%, the recall rate reached 70.84%. Experiment results show that the method can solve the problem of the fine-grained emotion classification from Chinese reviews with good effects.