Research on Link Prediction under the Structural Features of Attention Stream Network

The attention flow network is an important branch of network science, and the research on the link prediction algorithm of attention flow network has important theoretical and application value. The traditional link prediction algorithm based on the neighbor hypothesis cannot learn the global structure information of the network, while the traditional graph neural network algorithm needs to combine external information and has a large amount of calculation. Based on the structural characteristics and the advantages of the MLP neural network, this paper proposes a stru2vec-based algorithm. The link prediction SMLP algorithm is compared with the existing DeepWalk, Line, Node2vec, Hope, NetMF and other algorithms. It is found that the SMLP algorithm proposed in this paper improves on the four evaluation indicators of Precision, Recall, Accuracy and F1 by about 15.8% on average. The average increase in AUC index is about 12.1%, and the performance is better on the ROC curve, and the algorithm proposed in this paper focuses on learning the network structure, and has a good learning effect for the network with missing node label information and sparse self-characteristic attributes.