Community Detection In Heterogeneous Dynamic Network Based On Hypergraph Neural Networks

In the real world,the network has high complexity,which is embodied in the heterogeneity and dynamic evolution characteristics.The community characteristics of a complex network can reflect the local information of the data samples,which is abstracted by the network and the relationship between them.Studying the community in the network helps to understand the structure and function of the network.Traditional community discovery algorithms mostly study static and homogeneous networks,and cannot learn the complex relationships and dynamic evolution processes in the network.Therefore,research on community discovery algorithms suitable for heterogeneous dynamic networks has certain practical significance.This paper uses the network representation learning method to generate the embedded representation of the nodes in the network,and combines the hypergraph neural network(HGNN)to generate a community classification model for heterogeneous networks and a community discovery model for heterogeneous dynamic networks.The main contributions of this paper include:(1)The H?node2vec algorithm for heterogeneous networks is proposed.According to the structural characteristics of the heterogeneous network,this paper improves the walking strategy of the node2 vec algorithm,generates the feature vector and meta-path information of the node,and represents the node in the same dimension.Experiments show that compared with traditional representation learning methods,this algorithm has better embedding representation capabilities on heterogeneous networks.(2)An attention model fused with hypergraph is proposed(HAT).The meta-path generated by the heterogeneous network contains rich semantic information.In this paper,the concept of hypergraph is introduced to generate an incidence matrix based on the meta-path.Applying the incidence matrix to the graph attention model,the structure and semantic information between nodes in the meta-path can be learned at the same time.The experimental comparison between this model and the graph attention model before the improvement proves that it has better network representation capabilities on heterogeneous networks.(3)A community classification model for heterogeneous networks(Meta?HGAT)is proposed to realize community division on heterogeneous networks.This paper first improves the hypergraph neural network by adding meta-path information,so that the improved algorithm can learn the correlation between meta-paths.Then,this paper combines the above-mentioned H?node2vec and HAT model to design a community classification model based on hypergraph neural network(Meta?HGAT).The experimental comparison with the traditional heterogeneous network graph embedding algorithm proves that the model has better performance in community classification and community discovery tasks.(4)A community discovery model for heterogeneous dynamic networks(Meta?DHGAT)is proposed.Based on the proposed Meta?HGAT model,this paper further considers the time characteristics of the network,and generates a community discovery model for heterogeneous dynamic networks.Through the experimental comparison with the Meta?HGAT algorithm,it is proved that the improved algorithm has a better community division effect.