摘要
网络在线评论对于商家和顾客具有重要价值,因而日益受到虚假评论行为的冲击。作为两个重要的在线评论领域,产品类评论(如亚马逊、淘宝)和店铺类评论(如点评网、Yelp)在语言特性、评论行为等方面存在显著差异。虽然研究者们已提出大量针对产品类虚假评论的检测方法,但对于店铺类虚假评论的研究仍然较少。针对Yelp.com网站上旅店、饭店有标注的点评数据,提取并分析各种评论欺诈特征,利用多种有监督学习方法进行虚假评论检测。实验结果表明,检测精度最高可达74%,AUC值可达75%。虽然店铺类虚假评论具有极强的隐蔽性,但通过权衡检测精度和召回率,可利用有监督学习方法对店铺类虚假评论进行有效检测。
Due to the importance for both the merchants and customers,online reviews are increasingly under the attack of fake reviews.As the two main review domains,product reviews(e.g.Amazon,Taobao)and store reviews(e.g.Dianping.com,Yelp.com)significantly differentiate from each other in linguistics and behaviors.While product fake review detection attracts much research interests,store fake review detection has got less attention.In this paper,we focus on store fake review detection problem by exploiting the labeled datasets containing hotel and restaurant reviews from Yelp.com.Specifically,we extract and analyse a number of review spam features,with which we use supervised machine learning approaches to detect fake reviews.Experiments suggest that the maximum precision and AUC can reach 74%and 75%,respectively.Although the fake reviews from Yelp.com are very deceptive,supervised learning methods are effective in detecting fake store reviews by trading off detection precision and recall.
作者
王琢
汪浩
胡润龙
高珮
WANG Zhuo;WANG Hao;HU Run-long;GAO Pei(School of Information Science and Engineering,Shenyang Ligong University,Shenyang 110159,China)
出处
《软件导刊》
2020年第4期71-74,共4页
Software Guide
