基于同质多层图卷积的多尺度网络对齐模型  

Multiscale network alignment model based on convolution of homogeneous multilayer graphs

作　　者：崔家豪 江涛[1] 徐梦瑶 CUI Jiahao;JIANG Tao;XU Mengyao(Key Laboratory of Language and Cultural Computing of Ministry of Education,Northwest Minzu University,Lanzhou 730030,China)

机构地区：[1]西北民族大学语言与文化计算教育部重点实验室,甘肃兰州730030

出　　处：《智能科学与技术学报》2024年第4期522-532,共11页Chinese Journal of Intelligent Science and Technology

基　　金：中央高校基本科研业务费项目(No.31920240081,No.31920240127)。

摘　　要：社交网络对齐作为网络科学的重要研究方法,已在多个领域广泛应用。现有方法通常依赖高质量的用户属性信息来完成特定任务,但隐私保护机制的存在使得这些信息难以获取。此外,若仅依赖网络拓扑结构,可能面临数据不足的挑战。针对上述问题,基于节点邻域特征和网络同质性提出一种基于同质多层图卷积的多尺度网络对齐模型。节点特征方面,通过K近邻算法聚合节点邻域信息建模深层网络结构,从而进行数据增强。图卷积方面,根据网络同质性构建同质度矩阵,对卷积过程进行引导,并以网络社区结构为基础来处理不同尺度的社交网络。在两个不同规模的现实社交网络实验结果表明,该方法能够有效提升社交网络对齐任务的性能。Social network alignment as an important research method in network science has been widely used in several fields.Existing methods usually rely on high-quality user attribute information to complete specific tasks,but the existence of privacy protection mechanisms makes this information difficult to obtain.In addition,relying solely on network topolo‐gies can be challenged by insufficient data.In order to solve the above problems,a cross-network user alignment model based on the node neighborhood characteristics and network homogeneity was proposed.In terms of node characteristics,the K-nearest neighbor algorithm was used to aggregate node neighborhood information to model the deep network struc‐ture,so as to enhance the data.In terms of graph convolution,the convolution process was guided by the construction of a homogeneity matrix according to the network homogeneity,and the social networks of different scales were processed based on the network community structure.Experimental results on two real-world social networks of different scales show that the proposed method can effectively improve the performance of social network alignment tasks.

关 键 词：网络对齐 网络嵌入 网络同质性 多尺度网络 社区发现 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]