简介:一个网络的聚类的系数C,是在各种各样的节点的邻居之间的直接连接的一项措施,从0(为没有连接)到1(为完整的连接)。我们基于在从一个来源节点的距离h的节点定义一个小世界的网络的扩大聚类的系数C(h),因此概括distance-1邻居在计算平常的聚类系数C=C(1)采用了。在一个网络关于距离分发Pδ(h)基于已知的结果,也就是说概率顶点的随机选择的对有距离h,我们发源;试验性地验证法律Pδ(h)C(h)≤c木头N/N,在c是很少超过1的一个小常数的地方。因为它证明产品Pδ(h)C(h)由是比为Pδ(h)的最大的价值的产品更加小的价值是上面界限的,这结果是重要的;C(h)。聚类系数扩大;管理他们的法律提供新卓见进小世界的网络的结构;为他们的性质的进一步的探索开创大街。
简介:Inthispaper,weproposeadimension-reducing,K-meanclusteringprocedurebyProjectionPursuit(PP)techniquesoastoexploretheclusteringstructureofdatainhigh-dimensionalspaceintermsoflow-dimensionalprojectivepointsofdata,andweobtainthea.s.consistenceoftheestimatesoftheclustercentersandprojectionorientations.
简介:Mostoftheearlierworkonclusteringmainlyfocusedonnumericdatawhoseinherentgeometricpropertiescanbeexploitedtonaturallydefinedistancefunctionsbetweendatapoints.However,dataminingapplicationsfrequentlyinvolvemanydatasetsthatalsoconsistsofmixednumericandcategoricalattributes.Inthispaperwepresentaclusteringalgorithmwhichisbasedonthek-meansalgorithm.Thealgorithmclustersobjectswithnumericandcategoricalattributesinawaysimilartok-means.Theobjectsimilaritymeasureisderivedfrombothnumericandcategoricalattributes.Whenappliedtonumericdata,thealgorithmisidenticaltothek-means.Themainresultofthispaperistoprovideamethodtoupdatethe'clustercenters'ofclusteringobjectsdescribedbymixednumericandcategoricalattributesintheclusteringprocesstominimisetheclusteringcostfunction.Theclusteringperformanceofthealgorithmisdemonstratedwiththetwowellknowndatasets,namelycreditapprovalandabalonedatabases.