Research On Fake Review Detection Based On Graph Neural Network

In recent years,the development of e-commerce business has become more and more vigorous.When customers shop on e-commerce platforms every day,they also post a lot of reviews.Comments will affect users’purchase decisions,so many merchants on e-commerce platforms will publish many fake reviews under their peers and their own stores to wrongly influence users’ purchase decisions.In response to this problem,the traditional false comment detection algorithm is often to train a deep learning model for the feature of the comment text,but simply using the comment text as a feature cannot take advantage of the node interaction that exists widely in social networks.Low accuracy,frequent misjudgment and other problems.Aiming at the main problems of false comment detection in social networks,this paper improves and builds related algorithms from the perspectives of graph neural network and federated learning,and mines the interactive information and feature representation of nodes in the network in an all-round and multi-level manner,so as to accurately improve Performance of fake review detection algorithms.The following is the main work of this paper:(1)Aiming at the problem of low accuracy of detection methods based on comment text features in homogeneous social network graphs,we propose a fake comment detection model GFD based on graph neural network.According to the relationship between different types of nodes in the heterogeneous graph,the model extracts the homogeneous graph of comment nodes,and aggregates the characteristics of its neighbor nodes for the comment nodes in the homogeneous graph,and finally enhances the feature representation of the node itself.It has been verified by experiments that compared with the fake comment detection algorithm based on comment text features,the fake comment detection model GFD based on graph neural network can improve the accuracy of detection results.(2)Aiming at the problem of limited information in the singlemachine graph neural network model dataset,we propose a federated learning-based graph neural network fake review detection framework FGFD.The framework can unite multiple clients to learn feature representations of their entities in different dimensions in each other’s data set under the premise of protecting privacy,and let each participant cooperate without disclosing the underlying data and its encrypted form.modeling.At the same time,when the gradient is transferred between the client and the server,we introduce a differential privacy mechanism to ensure that the client data privacy will not be leaked in the gradient.Experiments have verified that compared with the stand-alone graph neural network model,the federated learning-based graph neural network false comment detection framework FGFD can improve the accuracy of detection results without revealing the privacy of the dataset.(3)From the perspective of actual project application,we have completed the construction of the false comment detection framework and algorithm implementation of the federated graph neural network,and completed the interface packaging to realize a false comment detection system based on the federated graph neural network.