Research On Heterogeneous Network Community Detection Algorithm Based On Graph Convolutional Network(GCN)

With the rapid development of artificial intelligence technology,many real complex systems can be abstracted as heterogeneous networks,such as social networks,transportation networks and protein networks.These networks have distinct community structures.The community detection task is dedicated to mining the community structure in networks,which contributes to a deeper understanding of the inner workings of complex systems.Furthermore,it can be applied to tasks such as social recommendation and link prediction,which has extremely important practical significance.In recent years,benefiting from the development of graph neural networks,community detection algorithms based on deep learning have become a popular research direction.Network representation learning is an important part of community detection,and existing heterogeneous network representation learning methods based on graph neural networks perform well.But there is still room for improvement in integrating the different semantics contained in multiple meta-paths.In addition,existing unsupervised community detection algorithms uniformly encode the redundant information in communities,which limits the potential of the methods to expand to larger-scale graphs.Meanwhile,the definition of community structure in heterogeneous networks is not yet unified,and it is not easy to unify when dealing with multiple types of nodes and multidimensional relationships.There is room for further exploration of community detection algorithms for heterogeneous networks based on graph neural networks.To address the above limitations,the main contributions of this paper are as follows:(1)Considering the limitations of existing heterogeneous network representation learning methods in exploiting higher-order indirect relationships between nodes,it is proposed that an embedding learning algorithm for heterogeneous network,named MGCN(Meta-Graph Convolutional Network).The algorithm includes two stages of heterogeneous adjacency matrices calculation based on meta-graph and learning node embedding.The heterogeneous adjacency matrix fuses different semantic information from multiple meta-paths and mines high-order indirect relationship between nodes.In addition,it can aggregate the neighborhood features of nodes into a unified pattern.This method reduces the embedding dimension,and then reduces the calculation time.Extensive experiments on public heterogeneous network datasets show that the proposed MGCN can outperform baselines and need less model training time in node classification and clustering tasks.(2)Considering the limitations of the existing unsupervised community detection algorithm in processing larger-scale graphs,it is proposed an enhanced contribution masked graph autoencoder for unsupervised community detection(ECCD).Firstly,the method introduces community contribution to measure the role played by nodes in maintaining the stability of the community structure.Then,identify the important nodes and marginal nodes in community.The ECCD masks the identified marginal nodes features and only uses a few important nodes to reconstruct the entire network features in an unsupervised way.Finally,using the learned valid node representations for community detection.Extensive experimental results on real-world networks indicate the effectiveness and the proposed method has more advantages in processing community detection in large-scale graphs.(3)Considering the problems of handling multiple types of nodes and multidimensional relationships in heterogeneous network community detection,it is proposed a multiple graph convolutional networkbased community detection algorithm for heterogeneous networks(GHCD).The method maps the features of different types of nodes into a unified feature space,and encodes the heterogeneous network using the MGCN.By reconstructing network features,GHCD learns the node representations that contain rich structural information and semantic information in heterogeneous network.Finally,the node representations are used for heterogeneous network community detection.The experimental results on real heterogeneous networks show that the proposed method can be used to divide communities containing different types of nodes.