基于多通道图卷积网络的节点聚类  

Node Clustering Based on Multi-channel Graph Convolutional Network

作　　者：孙艳丰 杜鹏飞 SUN Yanfeng;DU Pengfei(Faculty of Information Technology,Beijing University of Technology,Beijing 100124,China;Beijing Key Laboratory of Multimedia and Intelligent Software Technology,Beijing 100124,China)

机构地区：[1]北京工业大学信息学部,北京100124 [2]北京工业大学多媒体与智能软件技术北京市重点实验室,北京100124

出　　处：《北京工业大学学报》2023年第3期355-362,共8页Journal of Beijing University of Technology

基　　金：国家自然科学基金资助项目(62172023)。

摘　　要：针对在深度聚类中大部分基于图卷积网络(graph convolutional network,GCN)的方法仅使用拓扑图而忽略了特征空间中存在的结构信息的问题,提出一种通过引入特征图更充分地利用特征空间中存在的结构信息的节点聚类方法.首先,该方法使用自动编码器(auto-encoder,AE)来学习节点特征的潜在表示,同时在特征图、拓扑图及节点属性3个层面获得节点嵌入;然后,使用融合机制对学习到的节点嵌入进行融合;最后,通过自监督的方式训练网络实现节点聚类.在6个基准数据集上的大量实验表明,该方法明显提高了聚类精度.To solve the problem that most methods based on graph convolutional network(GCN)only use topological graph and ignore the structural information in the feature space in deep clustering,a node clustering method was proposed by introducing feature graph to make full use of the structural information in the feature space.First,an auto-encoder(AE)was used to learn the potential representation of node features,and the node embeddings were obtained at the three levels of feature graph,topology graph and node attribute at the same time.Then,a fusion mechanism was used to fuse the learned node embeddings.Finally,the network was trained by self-supervision to implement node clustering.A large number of expiments on six benchmark datasets show that the proposed method significantly improves the clustering accuracy.

关 键 词：节点聚类 图卷积网络(graph convolutional network GCN) 注意力机制 自动编码器(auto-encoder AE) 特征融合 图结构 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]