Research On Node Similarity And Classification Algorithm In Complex Network Based On GNN

Network node similarity is measured based on different similarity metrics to measure the similarity between nodes and is the basis of complex network research and application.As an important branch in the field of complex networks,node classification aims to identify the label categories of unknown nodes in the network.Many real-life problems can be transformed into node classification problems for study.However,the existing node similarity calculation methods cannot give the critical value of similarity,which makes it difficult to be applied in practice;the node classification model is difficult to train in the face of featureless networks,performs poorly on some topologically unbalanced graph data,cannot be closely connected with downstream tasks,and is difficult to be generalized to practical application scenarios.In view of the above problems,we propose a new similarity calculation method and node classification optimization model to improve the classification accuracy.At the same time,the node classification optimization model is transformed into network social layering problem for research by using the attention flow network dataset,which verifies the feasibility of the model.Finally,the model is applied in the practical scenario of high-speed railway station prediction.The main research contents are four aspects:Node similarity algorithm A-NFN in attention flow network based on GAT.First,Node2vec is used to do biased random walk to represent the node in the network as high-dimensional vector;second,NetworkX is used to construct the attention flow data into a graph;finally,all vector representations of nodes and the structure of attention flow network are taken as GAT model inputs to calculate the similarity between each pair of nodes.The method defines a critical value for similarity and extends the application of graph neural network algorithms to featureless networks.The experimental results show that the A-NFN algorithm proposed in this paper improves the accuracy of the results by an average of 1.6% compared with GCN.A preprocessing model Scatter-GNN is proposed for node classification in partial edge graphs.Firstly,the critical value of edge graph nodes is derived by the squeezing principle,and then the edge nodes in the graph are effectively identified;then,the potential neighbors of the edge nodes are represented by the calculation method of Hadamard product;then,attention mechanism is used to link all potential neighbors and their corresponding edge nodes;finally,the newly predicted graph structure and node vector are input into the GNN classifier to complete the training.Scatter-GNN does not bring noise to the original graph,and the model is evaluated on three public datasets and one private dataset.The accuracy is increased by 0.15,0.24,0.30,0.52 percentage points compared with GNN model.Network social stratification based on Scatter-GAT.Based on the online user behavior log,the click sequence is generated,and the similarity matrix of the click sequence between each individual user is calculated by the Jaccard coefficient,and the similarity network between users is constructed by combining the properties of the normal distribution function.Thus,the network social stratification problem is transformed into the classification problem of complex network under the edge graph structure,Scatter-GAT model based on graph attention network is proposed to solve the problem.The experiment shows that human behavior in the network can also be used to stratify the society.And the groups of the same educational level and income range have relatively consistent interests and hobbies.Scatter-GAT does not bring noise to the similarity network.Compared with the graph neural network model GCN,the average prediction accuracy is increased by 10.18%,and the layering recognition rate reaches92.47%.The high-speed railway station prediction model based on Scatter-GNN.Taking Yinchuan-Chongqing high-speed railway as an example,the high-speed railway station selection problem is transformed into the classification problem of edge graph structure under the complex network,and the Scatter-GNN model proposed is used to predict the station.After the network station is completed by the model,the new high-speed railway network and line label obtained are input into the GNN classifier to complete the station prediction.The results show that Baoji and Hanzhong are more likely to be the node stations in this north-south railway trunk line.The model can be used as an auxiliary strategy for the traditional route planning scheme,which may become a new way of studying such problems in the future.