Anomaly Detection Method Based Ongenerative Adversarial Graph Neural Network

With the rapid development of the Internet,people’s dependence on the network is increasing day by day,and the security and integrity of data is facing a severe test.Attributed network is a network composed of node attributes and edges connecting nodes.It is widely used in real life,such as social network,citation network,protein network,etc.The traditional attributed network anomaly detection method ignores the cross-modal interaction between the network structure and node attributes,and ignores the influence of anomaly features when the nodes interact.The emergence of graph neural network has brought a new direction for attribute network anomaly detection.Starting from the reconstruction of the network structure,the reconstruction error is used to judge anomaly nodes.This paper mainly studies the anomaly detection method of generative adversarial graph neural network.This paper firstly proposes an anomaly detection method for Generative Adversarial Graph Convolutional Networks for reconstruction network performance.The method improves the effect of the reconstructed network through the adversarial autoencoder,and the discriminator distinguishes the original network from the reconstructed network,and improves the efficiency of anomaly detection.The model is trained using the normalized attributed network anomaly detection dataset,and the node feature representation is obtained through the encoder.The generator uses the node feature representation to restore the reconstruction network,and the reconstructed network obtains the node feature representation through the encoder.The adversarial loss,reconstruction loss,and generator loss are introduced to optimize the model.The discriminator uses graph classification to distinguish the original network from the reconstructed network,and uses the adversarial autoencoder to improve the performance of the reconstructed network and node feature representation.The experimental results show that this method has better reconstruction effect and anomaly detection effect than only using graph convolutional network.This paper then proposes an anomaly detection method based on graph attention generative adversarial networks for the node representation learning problem.When a graph convolutional network aggregates node features,it is easy to aggregate abnormal features to normal nodes,and normal features to abnormal nodes.After the data is normalized with the model and algorithm based on the graph attention generative adversarial network,the multi-layer perceptron is used to reduce the dimension of the data,and the influence weight of the neighbor nodes on the node features is learned through the graph attention network,so as to aggregate the abnormal nodes Abnormal features,normal nodes aggregate normal features,highlight normal features and abnormal features.The discriminator adds the network feature representation to determine whether the input comes from the original network and the original network feature representation.The experimental results show that the method significantly improves the anomaly detection effect compared with the attribute network anomaly detection method without graph attention mechanism and without generative adversarial network.