Spammer Group Detection Based On Review Content And User Behavior

With the development of the Internet era,e-commerce has become an indispensable part of people's daily life.Faced with a large number of selectable products on major e-commerce websites,users will refer to the reviews of the product on the website before deciding to buy and take the experience of predecessors as the basis for whether to buy or not.Due to this background,some unscrupulous merchants have begun to hire some people to write fake reviews to brag about their products or disparage others' products to make themselves profitable,which will not only affect people's shopping experience but also destroy development of e-commerce industry.Therefore,it is of great significance to research the approach and system which can effectively detect fake reviewers for the e-commerce industry.The main research work of this paper is as follows:(1)This paper first proposes a method that combines commentsentiment and reliability called GSDCRS(Group Spammer Detection Combining Reliability and Sentiment)to detect these spammer groups.GSDCRS takes full consideration of the interrelationships among users,reviews and products,and combines the emotional consistency of users in the group.The algorithm first uses frequent itemset mining algorithm to generate some suspicious candidate groups,and then combining the relationship of user-product graph calculates the reliability of user,product and review,and mining emotional features based on spammer groups'emotional consistency.(2)Taking the consideration that the comment systems of most e-commerce websites now support comments with pictures,and the reviews with pictures often appear first when they are sorted,which makes the reviews with pictures more influential.Hence,this paper proposed a new method for detecting spammers called GSDIC(Group Spammer Detection Combing Images)based on graphic and text matching.This method mainly considers the matching degree of two dimensions,one is the matching degree of the picture and the product being evaluated,and the other is the matching degree of the picture and the review text.By training two neural networks to extract the main semantics of pictures and texts respectively,then calculate the similarity between the two semantics,and fuse these two kinds of features into the GSDCRS proposed previously,as to further improve the effect of spammer groups detection.(3)This paper built an e-commerce dataset based on JingDong,and verified the feasibility and effectiveness of GSDCRS algorithm and GSDIC algorithms on public Yelp dataset and JingDong e-commerce dataset.Experiments results show that the proposed method is superior to other machine learning algorithms.