融合评分-评价一致性和多维时间序列的虚假评论识别方法

Fake Comment Recognition Method Based on Comment-rating Consistency and Multi-dimensional Time Series

在线商品评论是指导客户购买行为的重要依据.虚假的评论无疑严重地误导消费者,为营造公平公正的在线购物环境带来巨大挑战.前人的研究主要分别依据评分或评论进行检测,一方面,研究成果没有斟辨评论内容与其评分的一致性,因为很多不真实的评论与其评分是不一致的;另一方面,研究成果没有考虑评论时间对识别虚假评论的影响,因为虚假评论使得一段时间内评论数量及评分会出现突变.因此,本文提出融合评分-评价一致性和多维时间序列的虚假评论识别方法.首先,根据在线商品评论的特点,分析评论文本的情感极性,判断与其评分的一致性;其次,考虑时间因素对评分及评论数量的影响,构建基于多维时间序列的虚假评论检测模型;最后,使用将抽取的多个虚假评论特征进行融合,构建出虚假评论检测分类器.通过实验对比验证了本文方法的有效性.

Online comments are critical important for customers online purchase.Fake reviews have seriously misled customers and further greatly challenged the fair and just online shopping environment.Studies prior about online goods comments were based on ratings or reviews separately.s.On one hand，they did not distinguish whether the contents of comments were matched with scores，as many fake reviews were not consistent with their ratings.On the other hand，the studies did not take into account the time impact on the fake reviews，since fake reviews often made the number of comments and its scores suddenly surge or decline in some time.Therefore，this paper proposed a fake comment recognition method based on comment-rating consistency and multidimensional time series.Firstly，based on the characteristics of online goods comments，we analyze the emotional polarity of the comments and judge the consistency of contents and ratings .Secondly，considering the influence of time factors on the number of comments and ratings，we construct an model to detect fake reviews based on multi-dimensional time series.Finally，we integrate all kinds of features of fake comments and train an effective fake reviews detection classifier.Compared with other state-of-art methods，our method in this paper works better.