一种基于加权多代表点的层次聚类算法.pdfVIP

计算机科学 2005Vo l. 32N2. 5 一种基于加权多代表点的层次聚类算法铃) 倪锥健 1 黄亚楼 1 事飞 2 刘赏 2 (南开大学软件学院 天津 300071)1 (南开大学信息技术科学学院 天津 300071)2 摘 要 CURE $.法是…种凝聚的层次聚~$.法，它首先提出了使用多代农点描述族的思想.本文通过对已有的基于 多代农点的层次采类算法特点的分析，提出了…种新的基于多代农点的层次我~界法 WRPC. 它使用了基于影响因 子的簇代农点选取机制和基于 k附近中~方法的小旅合并机制，可以发现形状、尺寸史为.1杂的革惠。实险练;良表明，该$. 法在保证执行效率的情况 r 取得了支好的来矣效果. 关键词层次聚巢，代农点 .k- 近~阂，数据挖掘 An Agglomerative Hierarchical Clustering Algorithm Based on Weighted Representative Points NI Wei-Jian1 HUANG Ya-Lou1 LI Fei2 L1U Shang2 (College of Software. Nankai University. Tianjin300071)1 (College of Information Technical Science. Nankai University ,Tianjin300071)z Abslract As an agglomerative hierarchical clustering algorithm. CURE firstly employs the method of representing clusters by selecting some representative points. Through the analysis of the feature of traditional hierarchical clus? tering algorithm. a novel agglomerative hierarchical clusterìng algorithm called WRPC ?s proposed in this paper. WR翩 PC can identiíy clusters with complex shapes and avrious size by introducing the influence 精 weight唰 based representative points selection mechanism and k-nearest糊 neighbor-method附 based clusters nesting mechanim. Experimenta! resu1ts show that WRPC can províde better clustering resu1t with high executing efficiency. Keywords Hierarchícal clustering , Representative points , k-nearest neighbor graph ,Data míning 法 WRPC (Weíghted Representatíve Points based Cluste‘ 1 ~I 宵