Research on Node Similarity in Attention Flow Network Based on GAT

The traditional mathematical model can not give the critical value of similarity in the study of node similarity in attention flow network, which makes its practical application difficult; For featureless networks, traditional deep learning methods are difficult to train. To solve the above two problems, A-NFN algorithm is proposed by using graph attention network GAT and graph embedding algorithm Node2vec. Firstly, the graph embedding algorithm Node2vec is used to perform biased random walk, and the nodes in the network are represented as high-dimensional vectors; Secondly, the attention flow data is constructed into a graph by using NetworkX toolkit; Finally, taking the vector representation of all nodes and the attention flow network structure as the input of the algorithm, the similarity between each pair of nodes is calculated, and the critical value of similarity is defined. Experiments show that the A-NFN algorithm proposed in this paper is compared with the popular graph neural network algorithm GCN in recent years, and the result accuracy is improved by 1.6 percentage points on average.