一种基于二部图谱划分的聚类集成方法  

A cluster ensemble approach based on bipartite spectral graph partitioning

作　　者：徐森[1] 皋军[1,2] 徐秀芳 花小朋[1] 徐静[1] 安晶[1] XU Sen;GAO Jun;XU Xiu-fang;HUA Xiao-peng;XU Jing;AN Jing(School of Information Engineering,Yancheng Institute of Technology,Yancheng 224001,China;Jiangsu Key Laboratory of Media Design and Software Technology(Jiangnan University),Wuxi 214122,China)

机构地区：[1]盐城工学院信息工程学院,江苏盐城224001 [2]江苏省媒体设计与软件技术重点实验室(江南大学),江苏无锡214122

出　　处：《控制与决策》2018年第12期2208-2212,共5页Control and Decision

基　　金：国家自然科学基金项目(61375001);江苏省自然科学基金项目(BK20151299);江苏省333工程项目;江苏省高等学校自然科学研究项目(18KJB520050);江苏省媒体设计与软件技术重点实验室开放课题项目(18ST0201)

摘　　要：将二部图模型引入聚类集成问题中,使用二部图模型同时建模对象集和超边集,充分挖掘潜藏在对象之间的相似度信息和超边提供的属性信息.设计正则化谱聚类算法解决二部图划分问题,在低维嵌入空间运行Kmeans++算法划分对象集,获得最终的聚类结果.在多组基准数据集上进行实验,实验结果表明所提出方法不仅能获得优越的结果,而且具有较高的运行效率.A bipartite graph model is brought into the cluster ensemble problem. The object set and hyperedge set are modeled simultaneously via a bipartite graph formulation considering the similarity among instances and the information provided by hyperedges collectively. A normalized spectral clustering algorithm is proposed to solve the bipartite graph partitioning problem, and the final clustering result is attained by performing K-means++ algorithm to partition object set embedded in low dimensional space. Experimental results on several baseline datasets show that the proposed approach is not only well-performed but also high efficient.

关 键 词：机器学习 聚类分析 二部图模型 聚类集成 谱聚类算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]