中文摘要：

在线产品评论中通常包含多个产品特征，影响评论接收者对产品的总体判断。表征产品整体、组件和性能的产品特征词之间存在各种关系，这些关系影响评论接收者的信息处理和整合过程。现有的评论倾向性合成方法主要考虑的是词汇在文本中出现的位置、频率等因素，没有考虑到词汇之间的关系。本文提出了一种基于产品特征词关系识别的倾向性合成方法，首先在产品特征词、情感词和修饰词提取的基础上，将特征句表示为一个三元组，然后识别产品特征词之间的等同、等级和相关关系，构建产品特征词关系图，在此基础上由底至顶、由类内至类间完成文本倾向性合成。我们在手机评论集上对该方法进行了验证，实验结果表明，使用该方法得到的评论倾向性值与在线评分、人工标注倾向性值之间存在显著的正相关关系，相关系数分别为0．701和0．823。

英文摘要：

Online product reviews, commonly containing multiple features of the product, influence the receiver＇s overall judgment on the product. There exist different relationships between product feature words, which have a significant effect on receivers＇ information processing and integration. Previous research concentrated on the locations and frequencies of the words and paid little attention to the relationship between the words. This paper proposed a method for combining online reviews sentiment orientation based on the recognition of relationship between product feature words, in which the relationship between product feature words was identified as synonymous, hyponymous and relevant, then corresponding relationship diagram was constructed. Based on the relationship diagram, text sentiment oriented combination was accomplished, following the order from bottom to top and from interior to exterior of the cluster. The method was verified on the dataset of mobile telephone reviews. Results indicate that there exist significantly positive relationship between the online reviews sentiment orientation values calculated by using the method, the online ratings and the manual tagged sentiment orientation values, the relative coefficients being as high as 0. 701 and 0. 823.