PageRank-based Review Spammer Group Detection

With the development of the Internet,the e-commerce platform in the form of online consumption has become the most popular mode of consumption nowadays.E-commerce platforms generally allow and encourage users to comment on products or services after they have been consumed,and user reviews have high value to both potential consumers and businesses.However,driven by interest,spammers mislead consumers by publishing false reviews to promote or discredit targeted.products.So detecting spam reviews is of great importance to both consumers and businesses.Research on the field of spam review recognition can be divided into spam reviews,spam reviewers and spam user groups.At present,the research mainly focuses on the spam review or the user's recognition,and has less attention on group detection.Traditionally,language features of the review texts or reviewers' behavior was used to detect spam reviews or users,but now due to the existence of a spam review filtering system,spam review users need to pretend to be real reviewers by manipulating their own behavior and review texts,Therefore,it is more difficult to detect spam reviews or users only by review texts or user behaviors.Moreover,a fake spammer user group(a group of reviewers collaborate to write fake reviews)is even more devastating because of the large number of people who have complete control over the evaluation sentiment of the target product.Therefore,detecting spam reviews is of great importance to both consumers and businesses.This paper finds that in the network structure consisting of review user groups,review users and products,spam messages can be transmitted in the review system network through the relationship among entities(groups,users and products are all called entities).This is similar to the fact that the web page node can propagate the authoritative value through the link relation in the web structure formed by the web page.Therefore,this paper proposes the GroupRank algorithm based on the PageRank,a classic algorithm based on the authoritative ranking of Web pages.When calculating a group's spam,not only the basic group spam but also the spam that the other nodes send to the group through the network are considered and weighted by the spam influence(the degree of the relationship between the entities),finally the adjustment factor is used to adjust the group based on the impact of spam and spam ratio.The experimental results show that GroupRank algorithm can effectively identify spam user groups,and the recognition effect is superior to the traditional ranking algorithm and classification method.