面向大规模数据快速聚类K-means算法的研究  

RESEARCH ON FAST CLUSTERING K-MEANS ALGORITHM FOR LARGE-SCALE DATA

作　　者：郭占元[1] 林涛[1] 

机构地区：[1]河北工业大学计算机科学与软件学院,天津300401

出　　处：《计算机应用与软件》2017年第5期43-47,53,共6页Computer Applications and Software

基　　金：天津市科技支持计划科技服务重大专项(14ZCDZGX00818)

摘　　要：为进一步提高K-means算法对大规模数据聚类的效率,结合MapReduce计算模型,提出一种先利用Hash函数进行样本抽取,再利用Pam算法获取初始中心的并行聚类方法。通过Hash函数抽取的样本能充分反映数据的统计特性,使用Pam算法获取初始聚类中心,改善了传统聚类算法依赖初始中心的问题。实验结果表明该算法有效提高了聚类质量和执行效率,适用于对大规模数据的聚类分析。To further enhance the efficiency of K-means clustering algorithm for large-scale data , combined withMapReduce computational model , a parallel clustering method is proposed , which uses Hash funand then obtains initial center by Pam algorithm. The sample extracted by Hash function can fully reflect the statisticalcharacteristics of the data , using Pam algorithm to obtain the initial clustering center , and improve the traditional clustering algorithm to rely on the initial center of the problem. It uses the Pam algorithm to obtain the initial clustering center , and improves the problem of that the traditional clustering algorithms rely on theresults show that the proposed algorithm can effectively improve the clustering quality and efficiency , and is suitable forthe clustering analysis of large-scale data.

关 键 词：大规模数据 聚类算法 MAPREDUCE Hash样本抽样 PAM算法 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]