中文摘要：

在现有褒贬性情感分类的研究中，缺乏对商品具体属性情感倾向的分析。基于此，建立细分类模型，将情感分类分为初分类和细分类两个过程。初分类确定商品评论的整体情感倾向，根据初分类的结果对商品的各个属性再次进行情感分类，以确定具体属性的情感倾向。从而消费者无需阅读具体的文本评论，就可以全面直观地了解商品，缩短做出购买决策的时间，降低决策的复杂度。该模型可作为网上商品销售的一个扩展功能使用，并利用酒店评论文本检测了模型的有效性。同时，论文通过对四种经典的特征算法的测试，发现在情感细分类中互信息（MutualInformation，MI）达到了更高的准确度。

英文摘要：

In present research of appraisable classification, there are almost no records of the sentiment inclination of the products attrib- utes. This paper constructs the concrete sentiment classification model by dividing the project into two processes： preliminary classification and concrete classification. The former will determine the overall sentiment orientation of reviews, the result of the process can be used for the further classification of each attribute of the product to determine the sentiment orientation of the product＇s each attribute. Thus, without reading comments, consumers can get a comprehensive and intuitive understanding of the product vividly. It helps shorten the time needed in making a buying decision and reduce the complexity involved. In this paper, the online hotel comments are used to Validate the model. The model can augment the function of the transaction virtual community. Meanwhile, in the validation test, the effect of the Mutual In- formation （ MI ） is found to be more accurate.