中文摘要：

作为情感倾向性分析的基础性工作，情感词典构建包括情感词的识别与极性判断两大任务。本文以亚马逊网站上的音乐商品评论信息作为数据源，力图构建该领域的情感词典。首先利用关联规则挖掘算法充分挖掘领域主题词和情感词之间的关系，获取体现领域特征的情感词；然后针对每个情感词，引入词项间的混合相关关系，结合PageRank模型构建情感词的量化图模型，获得每个情感词的极性。实验结果表明，本文所提方法能有效地构建音乐领域情感词典，不仅能够识别该领域特征的情感词，同时还能较为准确地判断该情感词的情感原极性。

英文摘要：

As fundamental work in sentiment analysis, opinion lexicon construction consists of two task, opinion word identification and orientation computing. This paper tries to build opinion lexicon based on music production review from amazon, com. Firstly, association rule mining algorithm is performed to mine the relation between field key word and opinion word, and opinion words are obtained. After that, the quantification model of opinion word is built to get original polarity, in which PageRank model combined with mixture relevance relation are introduced. Experimental Results show that the proposed method could construct opinion lexicon on music field effectively. It not only obtains the opinion word of field characteristics, but also provides more accurate judgment on their original opinion polarity.