Research On Graph-based Review Spammer Groups Detection Algorithm

Reviews on e-commerce platforms provide consumers with intuitive judgment of commodity status to a large extent.However,due to economic interests,many illegal merchants or professional shilling groups aim at the reviews on e-commerce platforms.They tried to influence consumers' judgments through fake reviews,so as to achieve the purpose of improving or reducing the sales of target commodities.This behavior impair the authenticity and fairness of e-commerce platforms seriously,major platforms have done a lot of work in users fake reviews to avoid the damage of fake reviews on the regular consumption order.Many experts and scholars have also put forward some solutions to reduce the harm of fake reviews.These solutions achieved some effect to a certain degree.However,the users of fake reviews gradually turn to organized fake reviews groups,many detection methods can't identify the fake groups well.In order to solve this problem,this paper makes an intensive study of the fake review group and puts forward two solutions.Firstly,through the use of mining a large number of user relationships method and discover the association between users.And then further find candidate groups.For each item,a user relationship graph is constructed by the user relation corresponding to an item and then a user relationship graph group was constructed.By mining the association strength of user relationship graph group,a number of sub-graph with high support users was founded as the candidate groups.Finally,we rank the association and tightness of candidate groups to find the fake reviews groups and users.In addition,based on the analysis of users' behavior,a fake review group detection method is proposed,which constructs user relationship graph based on the association among users and divides groups according to local expansion.Based on the undirected user relationship graph,by quantitative analysis of the relationship between users and apply the idea of local modularity,the relationship of the whole user relationship graph is segmented.The candidate groups was then founded.On this basis,the users' fake review behavior is analyzed and the characteristics of fake review groups are proposed.The suspicion degree of each candidate group is calculated according to the characteristics,and then the fake reviews groups and users was founded.Finally,the experimental results on Amazon_cn dataset and Yelp dataset are illustrated.By comparing with two existing algorithms,the effectiveness of the algorithm is verified.