Community Detection Fusing Graph Attention Network

It has become a tendency to use combination of autoencoders and graph neural networks for attribute clustering solve the community detection problem. However, existing methods do not consider influence differences between node neighborhood information high-order information, fusion structural features is insufficient. In order make better we propose model named fusing attention network (CDFG). Specifically, firstly an autoencoder learn features. Then only calculates weight on target but also adds After that, two are initially fused by balance parameter. The feature module extracts hidden layer representation calculate self-correlation matrix, which multiplied obtained preliminary achieve secondary fusion. Finally, self-supervision mechanism makes it face task. Experiments conducted six real datasets. Using four evaluation metrics, CDFG performs most datasets, especially with longer average paths diameters smaller coefficients.