Research On Review Spammer Groups Detection Based On Network Embedding

With the rapid development of e-commerce platforms,fraudulent comment group detection has become an indispensable part of maintaining platform security.At present,the detection methods mainly construct user relationships by considering the explosiveness of comments and the concentration of ratings.These methods often filter and screen the original data set,resulting in vector representations of users that can only capture part of the key information and cannot be clearly distinguished between vectors,which seriously affects the accuracy of detection results.Moreover,as fraudulent users become more adept at hiding their behavior,many existing detection methods capture insufficient structural and semantic information,which reduces the quality of candidate groups and has a great impact on the detection model.Some methods that detect fraudulent groups based on artificially designed indicators also greatly reduce the generality and robustness of the model.To solve these problems,this study conducts research from the following two aspects.Firstly,given the problem that existing methods make insufficient use of original data set information and lack of generality in artificially designed indicators,a fraudulent comment group detection method based on hypergraph embedding and autoencoder classifier is proposed.This method first makes full use of data set information to establish a hypergraph;then uses first-order and second-order similarity of hyper networks and fuses attention layers to obtain more representative vector representations of user nodes;then uses the Mean Shift method to cluster user vectors to obtain candidate groups;finally uses an autoencoder classifier with fusion linear transformation to detect the anomaly scores of each group,and sorts them to obtain fraudulent comment groups.Secondly,because of the problem that existing methods capture insufficient structural and semantic information and lack robustness,a fraudulent comment group detection method based on heterogeneous information networks and a teacher-student distillation network is proposed.This method first constructs a heterogeneous information network based on user comment behavior in an original data set;then uses an event embedding model to learn object embedding of heterogeneous information network to obtain vector representations of users;then uses the Mean Shift method to obtain candidate groups and learns vector representations of candidate groups using Paragraph2 vec model to obtain features of each group;finally inputs group features into the model for testing,take the difference between teacher network and student network as anomaly score,sorts them and obtains fraudulent comment groups.Finally,this paper conducts a large number of experiments on the Amazon Review Dataset,Yelp Miami Review Dataset,and Yelp New York Review Dataset,and compares them with existing methods to prove the effectiveness of the two methods.