基于分布式哈希表的分布式子空间聚类算法  

Distributed Hash table-based distributed subspace clustering algorithm

作　　者：曲琳[1] 周凡[1] 田翔[1] 陈耀武[1] 

机构地区：[1]浙江大学数字技术及仪器研究所,浙江杭州310027

出　　处：《浙江大学学报（工学版）》2010年第2期225-231,共7页Journal of Zhejiang University：Engineering Science

基　　金：国家"863"高技术研究发展计划资助项目(2003AA1Z2130);浙江省科技计划重大科技攻关资助项目(2005C11001-02)

摘　　要：提出一种基于分布式哈希表(DHT)的分布式子空间聚类(DISCLUS)算法,该算法对各结点存储的数据分别进行子空间聚类,对聚类结果进行合并,得到分布式系统的聚类结果.针对子空间聚类的特点,提出结果集缩减和结果集剪枝策略对结点间通讯进行优化.为实现结点聚类结果合并,提出分布式表决算法(DDV).该算法利用底层覆盖网的拓扑结构进行层次化表决信息收集,在动态网络环境中实现了对所有结点的无冗余覆盖.理论分析和实验表明,DISCLUS算法的聚类误差和通讯性能能够较好地适应系统数据集规模、网络规模和数据空间维度的增加.A distributed subspace clustering （DISCLUS） algorithm based on distributed Hash table（DHT） was proposed.Each node executed subspace clustering on its local data.Then the clustering results of nodes were combined to form the final clustering results of the distributed system.The dataset reducing and pruning schemes were proposed to optimize the communication between nodes according to the speciality of subspace clustering.A DHT-based distributed voting （DDV） algorithm was proposed to combine the clustering results of nodes.The algorithm used the topology of the underlying overlay to hierarchically collect the voting information.All the nodes in the system can be covered without redundancy.The theoretical and experimental results show that the clustering error and the communication cost of DISCLUS algorithm are scalable to the dataset,nodes and dimensionality.

关 键 词：对等网络 子空间聚类 分布式哈希表(DHT) 分布式表决 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]