生活消费平台虚假点评识别的研究与实现论文.docxVIP

生活消费平台虚假点评识别的研究与实现 摘　要 现今，以大众点评网为代表的生活消费平台已成为人们获取商家信息、反馈商家服务或产品质量的重要平台。与此同时，虚假点评作为一种夸大或诽谤目标商家口碑的商业行为在生活消费平台出现，并屡禁不止。虚假点评具有难以判断、数量众多的特点，并且对真实用户、诚信商家以及生活消费平台自身具有很强的危害性。因此，对于生活消费平台虚假点评的识别成为本文的主题。 虚假点评的识别在大体上可以视为一个二元分类问题，许多研究人员已经从文本分析及心理学，或者用户行为的角度进行了研究，取得了一定的效果。本文以大众点评网为例，在大众点评网获取的331,415条已标注点评的基础上对虚假点评展开研究。 本文总结前人经验，并在深入研究大众点评网上虚假点评的特征的基础上，从“可信度”的角度出发，分别提出了用户、商家可信度模型，并在此基础上对不同算法进行讨论，构建了虚假点评识别模型。本文同时提出了一套模型测试指标。经测试，模型AUC值达到0.9182481，表现优秀。最后针对模型的一些不足提出思考与展望。 关键词 生活消费平台 虚假点评识别 可信度  Fraud Reviews Detection on O2O ( D) platform ABSTRACT Nowadays, the Living-Consumption Platform (e.g. ) has become a very important Internet service for customers to view or submit comment on service or product quality, and for shopkeepers to improve their performance based on the feedback. At the same time, out of purposes of reputation exaggeration or defamation for some specific shop on the platform,fraud review, as a commercial activity, is very common among the platforms. Fraud reviews are extremely difficult to be detected, especially using the text analysis approaches.Also, fraud reviews are harmful for real users, honest businesses and platform itself. Therefore,a hybrid approach to detect fraud review is described in this paper. In general, identifyfraud reviews can be addressed as a binary classification problem, which has been studied by many researchers from the perspective of textual analysis and psychology, or user behavior to achieveresults. In this paper, the , from which 331,415 labeled reviews are derived, has been chosen for example to study the fraud review. This paper based its study on previous researches. With the in-depth study of feature of the fraud review on , the user-credibility and shop-credibility evaluation model wereraised respectively from the credibility perspective. Then different algorithms are discussed and a fraud review identification model is established. A reasonable evaluation method for the models is also proposed. The result shows that the model