基于领域特征指示词的隐式特征识别研究

Research on Implicit Feature Identification Based on Domain Feature Indicators

网络购物这一领域的迅猛发展带来了海量的在线评论数据,挖掘评论数据中所蕴藏的语义以及情感信息对用户以及商家都有着莫大的价值。在这样的应用需求背景下,出现了针对文本的情感分析(sentiment analysis)技术。但由于中文语言表达的多样性与复杂性,用户会在评论中含蓄地提到评价属性与观点。而现有研究对包含显式特征评论文本的情感分析已趋渐成熟,针对隐式评论句进行特征识别的却较少。因此,文中面向隐式特征识别这一研究难点,提出一种基于领域特征指示词的隐式特征识别方法。该方法首先利用构建的多词型的主题情感联合模型对特定领域内的显式评论句进行特征类别指示词的挖掘;再引入词向量模型作为衡量隐式评论句中线索词与特征指示词集中词项语义相关度的标准;最后分情形来实现对隐式评论句中线索词所属特征类别的指派。通过对不同产品的评论数据集进行实验,结果证明了该方法的有效性。

The rapid development of online shopping has brought a huge amount of online review data.The semantic and emotional information contained in the review data is of great value to both users and merchants.In this context of application requirements,sentiment analysis for text has emerged.Due to the diversity and complexity of Chinese language expression,users will implicitly mention evaluation attributes and opinions in comments.The methods of mining comments with display feature have become more and more mature,but the research on implicit feature identification is less.Therefore,an implicit feature identification method based on domain feature indicators is proposed for implicit feature identification.Firstly,the constructed multi-word thematic affective association model is used to mine the feature category indicators of the display comments in a specific field.The word2vec is used as a criterion to measure the semantic relevance between the clue word and the feature indicator word in implicit comments.Finally,the assignment of the characteristic category of the clue words in the implicit comment is realized by case analysis.The effectiveness of the proposed method is demonstrated by experiments on review data sets of different products.