基于“属性-情感词”汽车本体的文本情感分析-《计算机技术与发展》

文章信息/Info

Title:: Sentiment Analysis Method Based on Attribute-sentiment Ontology in Automobile Domain

作者:: 王连喜¹; 2; 1. 广州市非通用语种智能处理重点实验室,广东广州 510006; 2. 广东外语外贸大学信息科学与技术学院,广东广州 510006

Author(s):: WANG Lian-xi¹; 2; 1. Guangzhou Key Laboratory of Multilingual Intelligent Processing,Guangzhou 510006,China; 2. School of Information Science and Technology,Guangdong University of Foreign Studies,Guangzhou 510006,China

Keywords:: car reviews; Internet word-of-mouth; attribute-sentiment ontology; opinion sentence recognition; sentiment analysis

摘要:: 对特定领域网络评论进行情感分析,可以帮助商家更深入地了解用户需求、总结自身产品和服务的优势与不足,也可以帮助消费者了解特定领域产品各方面性能的评价分布,从而优化其消费决策。提出一种面向汽车领域的“属性-情感词”本体构建流程,并在此基础上提出基于“属性-情感词”本体的汽车评论文本观点句情感分析方法。该方法以观点句识别方法为基础,利用“属性-情感词”本体对汽车领域产品的八个维度(属性)进行情感分析,并与经典的朴素贝叶斯情感分类方法进行实验对比。结果表明提出的方法能有效提高属性层面上的情感分析准确率和召回率。但由于汽车领域的细粒度情感分析效果会受到“属性-情感词”本体的完善程度及相关规则的影响,因此需进一步完善“属性-情感词”本体,并构建更为全面的规则。

Abstract:: Sentiment analysis of reviews in specific areas can help businesses to better understand user needs,summarize the advantages and disadvantages of their own products and services,and also help consumers understand the distribution of performance of products in specific areas,thus optimizing their consumption decisions. We propose an ontology construction flow of “attribute-sentiment words”oriented to the automotive field, and propose sentiment analysis methods based on “attribute-sentiment words” ontology. Based on the viewpoint recognition method,this method uses the “attrib-ute-sentiment word” ontology to carry out the sentiment analysis of the eight dimensions(attributes) of the automotive products and compares it with the classical Naive Bayesian sentiment classification method. It is showed that this method can effectively improve the accuracy and recall rate at the attribute level. Since fine-grained sentiment analysis in the field of automobile is affected by the perfection of “attribute-sentiment word” ontology and relevant rules,it is necessary to further improve the ontology of “attribute-sentiment word” and construct more comprehensive rules.