Review Spammer Groups Detection Algorithms Based On Graph Neural Network

Consumers on e-commerce platforms usually make purchase decisions by referring to historical review information when purchasing goods.Therefore,many unscrupulous merchants will hire online supporters groups to make false reviews on goods,so as to improve their own product stars or reduce the reputation of their competitors.Compared with a single fake reviewer,fake review groups do more damage to the reliability of e-commerce platforms.Many researchers have studied fake comment group detection,but some existing detection methods can’t fully mine the interaction information between users and goods,so the neural network can’t fully learn the embedded representation of users,resulting in downstream learning can’t obtain good accuracy.In addition,some studies focus on manual construction of detection indicators,which is not only inefficient,but also difficult to design indicators to cope with different data sets.In view of the above problems,this paper proposes two solutions to fake comment group detection.Firstly,in order to solve the problem that the existing methods can’t fully explore the interactive relationship between users and goods,resulting in poor node embedding representation effect,we propose a group detection algorithm for fake comments based on graph convolution neural network.The algorithm constructs a layer attention convolution network model(LAGCN)to represent user feature vectors.LAGCN deeply mines the potential connections between users through graph convolutional network,and sums the embedding weights of different layers through attention mechanism,so as to represent the features of nodes in a more comprehensive and in-depth way.Then,the candidate group was obtained by clustering according to the obtained feature vectors,and the final fake comment group was obtained by ranking the candidate group through calculation of suspicion according to the indicators proposed in this paper.Secondly,a group detection algorithm based on semi-supervised graph neural network model is proposed to solve the problem that most research methods focus on constructing detection indexes through artificial feature engineering and the proposed indexes are difficult to cope with different data sets.The algorithm no longer relies on artificial detection indexes,but trains individual users with a few labels by constructing a semi-supervised graph neural network model,and learns to predict the labels of all users.And then through the obtained label to generate indicators,in order to replace manual extraction indicators for group detection.Finally,the two algorithms proposed in this paper are tested on Amazon data set and Miami data set,and compared with the existing four algorithms,to verify the effectiveness of the proposed methods.