基于力的类同传播聚类方法  

Force-Based Affinity Propagation Clustering Approach

作　　者：王士同[1] 陈晓峰[1] 曹苏群[1,2] 钟富礼[3] 

机构地区：[1]江南大学信息工程学院,江苏无锡214122 [2]淮阴工学院机械系,江苏淮安223001 [3]香港理工大学计算机系

出　　处：《江南大学学报（自然科学版）》2009年第4期379-385,共7页Joural of Jiangnan University (Natural Science Edition)　

基　　金：国家863计划项目(2007AA1Z158;2006AA10Z313);国家自然科学基金项目(60773206/F020106;60704047/F030304);2006年江苏省6大人才高峰计划项目;2008江苏省研究生创新计划项目

摘　　要：提出一种基于力的类同传播聚类方法AFAPC。依据万有引力定律,根据数据间的相互引力,并在数据组成的网络中交替传递类同信息,计算出数据的聚类和对应的聚类中心。实验结果表明,AFAPC能在更短的时间内,取得与类同传播聚类算法APC相媲美的聚类效果。Clustering is an important task in data processing community. Through a clustering processing, data are partitioned into groups according to their underlying distributions, and each cluster has a center. In this paper, an alternative force-based affinity propagation clustering approach called AFAPC is proposed. The clusters and the corresponding centers can be achieved by transforming affinity messages in data networks, where the affinity messages are obtained based on the gravity forces between data points. The experiments demonstrate that the clustering accuracy of the proposed approach AFAPC can be comparable with APC. However, its running time is much less than the affinity propagation clustering approach in [ 9]

关 键 词：聚类 类同传播 万有引力 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]