Heterogeneous Network Communitv Detection Based On WGAT

In the analysis task of heterogeneous networks,community detection is dedicated to discovering invisible communities in the network.Representation learning is an effective technology for community detection.Representation learning refers to obtaining network nodes through machine learning or deep learning,and then realizing some downstream work,such as clustering and node prediction.At present,graph convolution network is the most popular field in graph network learning.With its excellent modeling ability,its performance is far beyond the previous generation of deep learning model based on natural language processing.Although graph convolution network is developing rapidly,it is still difficult to deal with heterogeneous networks.It is clear that the current graph convolution neural network model can not be applied to heterogeneous network.At the same time the graph neural network requires the input of the global adjacency matrix of the graph network during calculating,so the high space-time complexity and computational overhead make it unable to be implemented in large-scale complex networks.Although the existing algorithms reduce the space-time complexity of graph convolution neural network,these algorithms still can not realize the convolution operation of heterogeneous network;on the other hand,the existing graph convolution neural network does not consider the importance of the neighbor nodes when aggregating the node neighbor characteristic information,which will make the important node information can not be saved;Finally,in the process of capturing high-level features of nodes,few algorithms consider the weight of edges between nodes in the network.To sum up,this paper proposes a graph attention network algorithm based on edge weight(WGAT).In order to capture the potential eigenvectors of heterogeneous network nodes and realize the community detection of heterogeneous network.This paper proposes a graph attention network algorithm WGAT(Weighted Graph Attention Network)based on edge weights to capture the advanced features of heterogeneous network nodes,and then realize the detection of heterogeneous network communities.The main work of this paper is as follows:(1)According to the target requirements,specify the meta-path to convert the heterogeneous network into the homogeneous network,which is convenient for the subsequent convolution operation.(2)Redefine the weight of edges between network nodes after homogenization according to the network structure information,so that the algorithm not only considers the importance of neighboring nodes,but also considers the importance of edges between nodes.