基于用户行为的产品垃圾评论者检测研究

Research on Product Review Spammer Detection Based on Users' Behavior

为找到垃圾评论的制造者,提出一种基于用户行为的产品垃圾评论者检测方法。从垃圾评论者的行为目的出发,将其发表垃圾评论的5种行为模式作为垃圾评论者的检测指标,从卓越亚马逊网站获取1 470个评论用户,按单指标选取、5个指标集成选取的方法确定最可能和最不可能成为垃圾评论者的评论用户各25个,并对这50个评论者进行人工标记,根据标记结果设计有监督的线性回归模型。实验结果表明,该模型从1 470个评论者中发现88个用户为垃圾评论者,对垃圾评论者的检测效果优于基于用户有用性投票的基准方法。

In order to find the review spammers,this paper proposes a user review spammer detecting method that is based on users＇ behavior.Starting from the purpose of review spammers,it makes spammer＇s five behavior patterns as the index of spammer detection.Based on the 1 470 reviewers,it gets from the JOYO Amazon website,according to a single indicator selection and 5 indicators integration selection,finally gets the 25 of most likely spammers and 25 of the most unlikely spammers,and markes artificially the 50 of suspicious reviewers,according to the artificial results trained a supervised linear regression model which is based on the 5 indicators.Experimental results show that the model of the spammer detection finds 88 review spammers in the 1 470 reviewers,and is more effective than those based on user voting usefulness of the baseline method.