基于重叠度的层次聚类算法.pdfVIP

计算机研究与发展 ISSN 1000-1239/CN 11-1777/TP Joumal of Computer Research and Development 44( Suppl. ): 181-186, 2007 基于置叠度的层次黯类算法 震中明阳俊大川举实软的青川山捕时公凋厦妒万门耸仔 1 AυAU 学训 句3ro ，到 mm A Hierarchical Clustering Algorithm ?ased on CφSharing Level Qu Jun , Jiang Qingshan, and Weng Fangfei (Softu地 re Schωl ， Xiamen University , Xiamen 361005) Abstract There are two critical problems for the clustering analysis techniques to select the appropriate value of number of clusters and partition overlapping and irregular data. Hierarchical clustering provides a good solution to them , but how to control the iterative process is the key. In this paper, a simple definition of cφsharing level is first given. Then, a fast hierarchical clustering algorithm based on it is proposed without specified number of clusters in advance, whose appropriate value can be decided in the iterative process. The algorithm stops clustering according to the cφsharing level between clusters. After discussing some related topics. real data are adopted to evaluate the p巳rformance of the proposed clustering algorithm and it is compared with other parallel algorithms. Key _words hierarchical clustering; co附 sharing level; overlap 摘要选择合适的聚类数和准确划分簇间 j竟是的数据是来类分析领域两个被广泛研究的问题.提出 了一个基于熏聋皮的层次聚类算法 (CCSLM) .该算法基于 1震聋皮的街景，而且不常要预先指定聚类数， 能够很好地解决以上两个问题.算法根据每两提之间的重叠情况自动运行或停止，从而准确划分簇间 重叠的数据，并自动确定最伎聚类数.通过周像分割应用表明算法的有效性和鲁棒性.还把它和其他聚 类算法进行比较和分析.以证明该算法的优越性. 关键诩层次聚类;重叠度;重叠 中图法分类号丁P311 .13 应次聚类算法在准确划分类问重叠的数据或者 的数据聚类效果较好，但是计算比较复杂. 是不规则数据方面有较好的表现.通常，居次聚类 本文提出一个算法，克服传统层次聚类算法在 的过程由分裂和合并两个子过程组成.日前，人们 划分族问重叠的数据时放果较差的不足;同时具有 巳经研究出很多的层次聚类算法，如 ISODATA[I] ,