融合网络结构和属性嵌入的社区发现方法  

Community Discovery Method Integrating Embedded Network Structure and Attributes

作　　者：吴琪 黄芳[1] 王淑勤 卢港 WU Qi;HUANG Fang;WANG Shu-qin;LU Gang(School of Computer Science and Engineering,Central South University,Changsha 410083,China)

机构地区：[1]中南大学计算机学院,长沙410083

出　　处：《小型微型计算机系统》2023年第7期1542-1549,共8页Journal of Chinese Computer Systems

基　　金：湖南省科技计划项目(2016JC2011)资助。

摘　　要：在社区发现中,针对不同的应用场景和网络中愈加丰富的节点属性信息,目前缺乏一种普适性的社区发现方法.本文提出了一种基于综合社区性嵌入(Comprehensive Community Embedding,CCE)的社区发现方法.CCE基于节点的社区相似性假设,进行社区游走,捕获网络结构的邻接特征、局部模块度和边介数的社区特征,并结合节点属性这一辅助社区特征,实现节点的综合社区性表征;社区划分通过K-聚类神经网络模型(K-Clustering Neural Network Model,K-CNN)以无监督节点聚类的方式实现.在社交网络和合著网络两类网络数据集上的实验结果表明,本文方法(CCE+K-CNN)表现稳定且优于传统方法,CCE采用的社区游走方法较其他游走方法更适用于解决社区发现问题.In community discovery,there is currently a lack of a universal community discovery method because of the different application scenarios and the increasingly abundant node attribute information in the network.This paper proposes a community discovery method based on Comprehensive Community Embedding(CCE).Based on the community similarity assumption of nodes,CCE performs CommunityWalk to captures the community feature of the network structure′s adjacency,local modularity and edge betweenness,and realize the comprehensive community characterization of nodes with combining node attributes as an auxiliary community feature;community division is achieved by clustering with unsupervised nodes based on the K-clustering neural network model(K-CNN).Experimental results on two types of network data sets of social networks and co-authored networks show that the method in this paper(CCE+K-CNN)is stable and superior to traditional methods.The CommunityWalk method adopted by CCE is more suitable for solving community discovery problems than other walking methods.

关 键 词：社区发现 综合社区特征 社区性游走 聚类神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]