Research On Review Spammer Detection Based On Review Relationship

Review spammer detection have played an important role whether in the recommendation algorithm or in opinion mining.At present,the identification on review spammer detection has attracted lots of attention from scholars both domestic and oversea,and a lot of researches have achieved.But those achievements base on review relationship have not mentioned the relationships among reviewers,and they also haven't mentioned the detection method,which combined characteristic of the comment with the relationships among reviewers to identify review spammer.Therefore,we will focus on the identification of the product review spammer in this paper and the details of the research are proposed as follows:(1)Presently,the research of review spammer detection mainly is based on relationships among reviewers,reviews and stores,which doesn't take the relationships among reviewers into consideration.This paper proposes a multi-edge graph model to identify review spammer.Firstly,in the multi-edge graph model,the nodes represent reviewers and the edges represent the relationships among reviewers.Secondly,according to multi-edge graph model,reviewers' inter-assess trustiness model is based on PageRank algorithm to identify review spammer.And lastly,the datasets are crawled from JOYO Amazon website and Resellerrating.com.The experimental results show that the model can achieve better performance on the accuracy of review spammer detection and the identification of review spammer who had only one review can be solved in some extent.The accuracy of this method has been improved by 13%and 14%.(2)The approach applies the PageRank algorithm to calculate each reviewers'inter-assess trustiness,but those review spammers might use linking cheating way to improve their rankings,This paper proposes the model is based on trust propagation algorithm to identify review spammer.Meanwhile,this algorithm tries to establish the relationship among reviewers first,and then establish the trustiness model considering TrustRank and review honesty.The method of random walk process on the graph model can calculate reviewers' trustiness.At last,with the reviewers' trustiness,review spammers can be detected.Experimental results according to two datasets indicate that the accuracy of this method has been improved by 2%compared with the approach from the reviewers' inter-assess trustiness model.(3)On account of the problem appear in trust propagation approach to identify review spammer,which fail to make use of reviewers' distrust propagation in the process of spammer detection,we propose a kind of detection algorithm combining TrustRank algorithm with Anti-TrustRank algorithm.First,this algorithm base on the reviewers' relationship graph,and consider each reviewer as a joint.When we add the conflicting relationships among reviewers,the complex graph was finished.After that,we use Anti-TrustRank algorithm and TrustRank algorithm random walk respectively on conflicting edge and supportive edge of the graph model.Then we get the reviewer's distrust score and trustiness score.Finally the linear combination model combine TrustRank algorithm with Anti-Trustrank algorithm to identify review spammer.Experimental results according to two datasets indicate that the accuracy of this method has been improved by 2%compared with the model that fail to make use of reviewers' distrust propagation,and our method has been improved by 17%and 18%compared with baseline method.