AGCFN:基于图神经网络多层网络社团检测模型.

Multiplex network community detection methods based on graph neural network face two main challenges. Firstly, how to effectively utilize the node content information of multiplex network; and secondly, how to effectively utilize the interlayer relationships in multiplex networks. Therefore, this paper proposed the multiplex network community detection model AGCFN. Firstly, the autoencoder independently extracted the node content information of each network layer and passed the extracted node content information to the graph autoencoder for fusing the node content information of the current network layer with the topology information through the transfer operator to obtain the representation of each node of the current network layer, which made full use of the node content information of the network and the topology information of the network. The modularity maximization module and graph decoder optimized the obtained node representation. Secondly, the multilayer information fusion module fused the node representations extracted from each network layer to obtain a comprehensive representation of each node. Finally, the model under went training, and it achieved community detection results through a self-training mechanism. Comparison with six models on three datasets demonstrate improvements in both ACC and NMI evaluation metrics, thereby validating the effectiveness of AGCFN. [ABSTRACT FROM AUTHOR]