Research On Spam Review Detection Based On Graph Neural Networks

The growth of the internet has led to a shift towards online consumption and services.Consumers read and write reviews on public websites and e-commerce platforms to inform their purchasing decisions.However,the openness of these platforms has allowed for the emergence of the gray industry of posting spam reviews intended to mislead consumers,thereby enhancing product image or undermining competitors.These reviews harm consumers and disrupt e-commerce.This paper proposes using a graph neural network method to detect spam reviews based on behavioral and linguistic features in the review data.The main research contents are as follows:(1)The improved Att＿Graph SAGE model which introduced the attention mechanism on the isomorphic graph was proposed.This model assigns weights to the neighboring nodes of each central node based on the attention between nodes,samples and aggregates the neighboring nodes according to the weights,realizing the improvement from equal treatment of all neighbor nodes to different attention to neighbor nodes of different importance.A multi-view approach is used to model heterogeneous review networks of multiple composite relations.By modeling multiple composite relationship heterogeneous comment networks through a multi-view approach,the method learns node representations of comment nodes in each relationship on the isomorphic single view,and then combines the views using attention to obtain representation vectors of nodes with rich information.Experiments on real review data sets show that the Att＿Graph SAGE method with multiple views has better performance in identifying spam reviews.(2)Aiming at the category imbalance problem of fake review data,combined with boosting ensemble learning framework,this paper uses muti-view based Att＿Graph SAGE as Adaboost base learner to deal with unbalanced data sets,and trained the Att＿Graph SAGE classifier by serialization.The errors of the previous Att＿Graph SAGE classifier are used to update the sample weights of the next Att＿Graph SAGE classifier and applied in the cross entropy error function of the back propagation learning algorithm to improve the performance.The transfer learning strategy is used to accelerate the training of the model.By comparing Att＿Graph SAGE before and after boosting integration,ensemble learning can improve the performance of Att＿Graph SAGE on unbalanced data.