一种高效的K-medoids聚类算法  

Efficient K-medoids clustering algorithm

作　　者：夏宁霞[1] 苏一丹[1] 覃希[1,2] 

机构地区：[1]广西大学计算机与电子信息学院,南宁530004 [2]广西工学院计算机工程系,广西柳州545006

出　　处：《计算机应用研究》2010年第12期4517-4519,共3页Application Research of Computers

摘　　要：针对K-medoids算法初始中心点选择敏感、大数据集聚类应用中性能低下等缺点,提出一个基于初始中心微调与增量中心候选集的改进K-medoids算法。新算法以微调方式优化初始中心,以中心候选集逐步扩展的方式来降低中心轮换的计算复杂性。实验结果表明,相对于传统的K-medoids算法,新算法可以提高聚类质量,有效缩短计算时间。Due to the disadvantages of sensitivity to the initial selection of the medoids and poor performance in large data set processing in the K-medoids clustering algorithm,this paper proposed an improved K-medoids algorithm based on a fine-tuned of initial medoids and an incremental candidate set of medoids. The proposed algorithm optimized initial medoids by fine-tuning and reduced computational complexity of medoids substitution through expanding medoids candidate set gradually. Expen-rimental results demonstrate the effectiveness of this algorithm,which can improve clustering quality and significantly shorten the time in calculation compared with the traditional K-medoids algorithm.

关 键 词：聚类 K-medoids算法 中心微调 增量候选 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]