基于属性重要性的加权聚类融合  

Weighted Cluster Ensemble Based on Significance of Attribute

作　　者：阳琳赟[1] 周海京[1] 卓晴[1] 王文渊[1] 

机构地区：[1]清华大学自动化系,北京100084

出　　处：《计算机科学》2009年第4期243-245,249,共4页Computer Science

摘　　要：聚类融合是数据挖掘研究的一个热点。当前相关研究大多没有考虑进行融合的聚类成员的质量,因此较差的成员和噪声会对融合结果产生不良的影响。提出了一种对聚类成员进行加权的融合方法。该方法引入粗糙集理论中的属性重要性度量,根据聚类成员对融合的重要性赋予其权重,生成加权共生矩阵,进而产生融合结果。实验结果表明,提出的方法能较好地处理聚类成员间的质量差异,并能有效地消减噪声对融合的影响,从而得到更好的聚类融合结果。Cluster ensemble is a hot topic in data mining research. Resent research mostly pays little attention to the qualities of cluster members. However, bad cluster members and noise may affect the ensemble result. A weighted cluster ensemble approach was proposed. This approach set weights to all cluster members according to the significance of them relative to the ensemble result. The significance of each cluster member was evaluated through information measures of significance of attribute in rough set theory. Then weighted co-association matrix was generated and the final ensemble result was obtained. The experimental results show that the proposed approach can handle well different-quality of cluster members and lessen the affect of noise effectively. Therefore,it can afford better ensemble result compared with general cluster ensemble methods.

关 键 词：聚类融合 共生矩阵 属性重要性度量 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]