A Spatio-Temporal Tensor Graph Neural Network-Based Method for Node-Link Prediction in Port Networks

Port network information security has received extensive attention in recent years, in which the prediction of node links in the network is significant. A Port network is a dynamic network, and its structure changes continuously with time. Therefore, to effectively utilize the information of the dynamic network and improve the prediction efficiency as well as the prediction accuracy, this paper proposes a spatio-temporal tensor graph neural network model, which learns graph structural features from both spatial and temporal aspects to capture the evolution of the dynamic network. Firstly, the model extracts the corresponding spatial features for the port network node links using the graph compression technique and updates the spatial features based on GraphSAGE by sampling and aggregating the neighbor node features. Secondly, for the port network temporal links, this paper introduces an exponential decay model to extract the corresponding temporal features and obtains the temporal edge embedding vectors by convolution method based on the GCN model. The updated spatial features and temporal edge embedding vectors are then weighted for feature fusion to predict the security of port network links. Finally, this paper conducts experiments on seven publicly available dynamic graph datasets, and the results show that the prediction accuracy of the spatio-temporal tensor graph neural network model is better than the baseline models, such as GC-LSTM, HTGN, and GAT, in general.