基于词共现和词上下文的领域观点词抽取方法  

Domain opinion words extraction method based on word co-occurrence and word context

作　　者：宋施恩 樊兴华[1] 

机构地区：[1]重庆邮电大学计算机科学与技术学院,重庆400065

出　　处：《计算机工程与设计》2013年第11期4012-4015,共4页Computer Engineering and Design

基　　金：重庆市自然科学基金计划基金项目(CSTC;2009BB2079)

摘　　要：为提高领域观点词的抽取效果,主要研究了词共现和词上下文对领域观点词抽取的影响。引入词上下文生成同义词词表的方法,使用词上下文构造的向量表示该词语,考察词集与种子词语向量间的相似度,完成观点词的抽取和判别。提出了一种组合词上下文与传统考虑词共现的SO-PMI(senmantic orientation-pointwise mutal information)方法。实验结果表明,该方法有一定效果,相较于SO-PMI在性能上有较大提高,从一定程度解决了领域适用性的问题。To improve the effect of the domain opinion word extraction, the methods of word co-occurrence and word context are studied. The synonym vocabulary generation method of word context is referenced, then, a consideration of and word context field perspective word extraction method is presented. Word context constructed vectors are used to represent the word and the similarity between the words vector, which is calculated to extract opinion words. The SO-PMI method is improved by combining the above methods. The experiments show that in a certain extent, this method, compared to the SO-PMI, solves the problem of the dependence on domain.

关 键 词：领域观点词抽取 词共现 词上下文 倾向性判别 SDO-PMI 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术]