网络商品评论的特征–情感词本体构建与情感分析方法研究

Research on Construction of Feature-Sentiment Ontology and Sentiment Analysis

【目的】解决情感分析领域使用通用情感词典进行情感分析时,在特定领域内无法识别领域专用情感词,以及同一情感词描述不同特征时可能表达出不同情感倾向的两个问题。【方法】提出一种基于领域专用情感词的网络评论情感分析方法。该方法构建特征–情感词本体,利用本体对网络上的产品评论进行情感分析。并与基于Senti-HowNet词典的情感分析方法进行对比。【结果】本文方法在特征层的情感倾向分析的准确率和召回率都有显著提高。【局限】本文方法中的本体需要尽可能完整的特征词集和情感词集,并且情感分析结果好坏直接依赖于本体的构建是否完善;由于网络文本的不规范性,特征词和情感词抽取以及情感分析的过程都不考虑句法结构;数据分析过程对问题进行了简化,仅考虑特征粒度的情感倾向,未考虑连词等对情感倾向有影响的其他因素。【结论】对专用情感词和通用情感词进行分类管理,解决了两个问题,情感分析结果得到提高。

[Objective] In a specific domain, sentiment analysis, mostly based on general lexicon, cannot identify the context-specific sentiment belonging to the domain. Also, the same word in the specific domain shows different polarities （positive, negative, neutral） when describing different properties. The objective of this paper is to solve the problems described above. [Methods] A sentiment analysis approach based on domain-oriented specific sentiment phrases is proposed. By developing feature-sentiment Ontology, general sentiment and specific sentiment can be divided during the process of sentiment analysis. [Results] The proposed method shows fairly better results of precision and recall in terms of phrase-level sentiment analysis. [Limitations] In order to get better analysis, the Ontology should cover the concepts in the related field as much as possible and should be well-built; the authors ignore the syntactic rules during the concept extraction and sentiment analysis, because the product comments are not normative; in the phase of sentiment analysis, the authors assume that the context like conjunction would not affect the polarity. [Conclusions] The new method not only makes improvement on sentiment analysis by solving the problem described above, but also proposes a new way for sentiment lexicon management.