Research On Bipartite Graph Based Review Spammer Group Detection Method

At present,spammer review behaviors in e-commerce platforms are gradually changing from individual behaviors to group behaviors.Group behavior is a group of users who make spammer review on the target product through mutual cooperation.Compared with individual behavior,group behavior is more destructive and will cause greater harm to the e-commerce platform.The current detection method for spammer review groups mainly adopts the method based on frequent item mining,but can only detect tightly coupled groups;Or adopts the detection method based on the review graph structure.However,only directly related features can be extracted,and the ability to extract indirect related features is limited,resulting in insufficient detection accuracy.This paper has conducted some effective research on the problem that the existing spammer group detection methods have insufficient detection accuracy and cannot extract more accurate indirect association features.First,in order to solve the problem that the existing spammer group detection methods lack the extraction of indirect user-related features,which leads to a decrease in detection accuracy,this paper studies bipartite graph embedding to detect spammer groups.First,construct a bipartite graph of user product relationships based on the relevance of reviews;then,use the bipartite graph embedded network representation learning technology to learn the review characteristics of each user node in the bipartite graph of user product relationships,and mine the indirect association between users features.And generate a review feature vector according to each user’s review feature;Then,use a clustering algorithm to cluster each user’s review feature vector,and divide users with relatively similar review features into the same group,there by generating candidate groups;Finally,the unique features of the spammer group are used to generate corresponding detection indicators to identify candidate groups,and the detection of the spammer group is completed.Secondly,the traditional method of spammer group detection based on spammer behavior requires frequent item mining of review data,so this type of method requires a certain degree of closeness to the detection dataset,and can only detect ordinary real data set detection tightly coupled groups.Therefore,this type of method has certain limitations in detection accuracy.In order to break through this limitation,this paper studies the decomposition of bipartite graphs.First,the user product relationship bipartite graph is also constructed based on the review association.Then,the user relationship bipartite graph is decomposed to decompose the user relationship isomorphism graph;and then use the isomorphism graph embedded network representation learning technology to generate a user review behavior portrait for each user’s review behavior.Based on user review behavior portraits and using K-means++ clustering algorithm,users with similar review feature portraits are divided into the same candidate group;finally,multiple spammer group features are used to generate detection indicators to identify candidate groups.Thus,a spammer group is detected.Finally,in order to verify the effectiveness of the method proposed in this paper,experiments were conducted on the AMAZON and Yelp NY datasets and compared with some existing methods for detecting spammer groups.