基于双视角网络嵌入聚类集成社区发现算法  

Community detection algorithm based on dual-view network embedded clustering integration

作　　者：王英楠 郑文萍[2,3] 杨贵 WANG Yingnan;ZHENG Wenping;YANG Gui(Fenyang College of Shanxi Medical University,Fenyang 032200,Shanxi,China;School of Computer and Information Technology,Shanxi University,Taiyuan 030006,Shanxi,China;Key Laboratory of Computation Intelligence and Chinese Information Processing of Ministry of Education,Shanxi University,Taiyuan 030006,Shanxi,China)

机构地区：[1]山西医科大学汾阳学院卫生信息管理系,山西汾阳032200 [2]山西大学计算机与信息技术学院,山西太原030006 [3]计算智能与中文信息处理教育部重点实验室(山西大学),山西太原030006

出　　处：《山东大学学报(工学版)》2025年第1期41-50,共10页Journal of Shandong University(Engineering Science)

基　　金：国家自然科学基金资助项目(62072292);山西省1331工程资助项目。

摘　　要：针对现有网络嵌入方法忽略高阶结构,嵌入过程与社区发现任务独立进行,影响社区发现质量的问题,提出基于双视角网络嵌入聚类集成社区发现算法(community detection algorithm based on dual-view network embedded clustering integration,DNECI),算法包括双视角网络嵌入和聚类集成两部分。双视角网络嵌入模块对网络属性信息与拓扑信息实现自适应融合,保留网络属性信息与拓扑的高阶结构。聚类集成模块包括模块度优化和聚类优化两个组件,模块度优化组件利用高阶拓扑结构得到具有最优模块度的社区结果;聚类优化组件通过自监督聚类方法在嵌入空间得到聚类结果;引入互监督机制使两种视角的社区发现结果具有一致性。在4个真实数据集与15个算法进行对比试验,结果表明,DNECI在准确率和标准互信息至少比最先进的基准算法平均提高2.5%和1.4%,在调整兰德系数和F1分数至少平均提高3.7%和1.7%,具有较好的社区发现效果。In response to the issues where existing network embedding methods neglected higher-order structures,and the embedding process was conducted independently of the community detection task,which affected the quality of community detection,a community detection algorithm based on dual-view network embedded clustering integration(DNECI)was proposed.The algorithm consisted of two parts:dual-view network embedding and clustering integration.The dual-view network embedding module adaptively fused network attribute information with topological information,preserving the higher-order structures of both.The clustering integration module included two components:modularity optimization and clustering optimization.The modularity optimization component used higher-order topological structures to achieve community results with optimal modularity,while the clustering optimization component obtained clustering results in the embedding space through a self-supervised clustering method.A mutual supervision mechanism was introduced to ensure consistency between the community detection results from both perspectives.Comparative experiments on 4 real datasets and 15 algorithms showed that DNECI improved accuracy and normalized mutual information by at least 2.5%and 1.4%on average compared to state-of-the-art benchmark algorithms,and improved the adjusted Rand index and F1 score by at least 3.7%and 1.7%on average,demonstrating better community detection performance.

关 键 词：社区发现 网络嵌入 模块度 自监督 高阶结构 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]