Research On Community Detection Method Based On Graph Neural Network

With the arrival of the information era,various kinds of networks have been integrated into all aspects of people’s lives,such as interpersonal networks and dissertation citation networks,which are characterized by complexity and diversity.Community detection,on the other hand,refers to finding community structures with similar characteristics in network graphs to understand their topology and attribute information,so that they can be applied to tasks such as classification and prediction to serve the real society,and therefore community detection is of great practical significance.As the emergence of complex networks,the networks not only have a large number of nodes,but also contain diverse node features with important attribute information.This poses a challenge to the traditional community discovery methods,which basically deal with the structural information of the graph without fully exploring the content in the attribute information.These methods have achieved good results on networks without node features,but when facing the datasets of today’s large networks,a new method is needed to learn both the topology and node features of the network.Graph neural networks are the application and innovation of traditional deep learning methods on graph structured data for extracting feature representations in graphs,and the emergence of this technique makes up for the shortcomings of traditional methods.Therefore,this paper proposes a community detection algorithm(DGAET)based on node importance and double autoencoders,which learns both the structural and attribute information of the network and implements the weighted summation of features to finally achieve the division of communities.The main work of this paper is as follows.First,the research background and significance in the field of community discovery are described,and the existing community detection methods are categorized and summarized,followed by the theoretical background required to implement the algorithms in this paper,such as relevant techniques in deep learning and graph neural networks,for theoretical preparation.Second,based on the node importance,the original matrix is reconstructed into a node similarity matrix,as a way to highlight the structural information of the network and facilitate subsequent feature extraction.And the attention mechanism-based graph autoencoder is used to learn the structural information of the network,and the deep autoencoder learns the attribute information of the network,which realizes the full exploitation and utilization of the information in the graph.Third,multiple sets of comparison experiments are conducted to evaluate the performance of the DGAET algorithm on each real dataset.The results demonstrate that the algorithm also possesses higher accuracy than other existing graph neural network algorithms as the number of nodes and features increases.