聚类融合方法综述  

Clustering Ensemble Approaches: An Overview

作　　者：阳琳贇 王文渊[1] 

机构地区：[1]清华大学自动化系,北京100084

出　　处：《计算机应用研究》2005年第12期8-10,14,共4页Application Research of Computers

摘　　要：在分类算法和回归模型中,正广泛而且成功地使用着融合方法,该方法能克服分类、回归中的不稳定性,并给出较好的结果。在非监督机器学习领域,由于缺乏数据集的先验知识,所以分类和回归中的融合方法就不能直接用于聚类算法,这导致了该领域中对融合方法研究的起步较晚;近几年的研究和实验表明,聚类融合方法能很好地提高聚类算法的鲁棒性和稳定性。对近年来聚类融合的方法进行了综述,阐述了近年来对聚类融合方法进行研究的主要内容与特点,并讨论了聚类融合方法的研究方向。Ensemble approaches are widely and successfully used in classification algorithms and regression models. It can offer better results for overcoming instabilities in classification algorithms and regression models. However, in unsupervised learning, the researches of ensemble approaches are concerned only in recent years. Because the prior information of data sets in unsupervised learning is unknown, the ensemble approaches of classification algorithms and regression models can＇ t be utilized in the same way directly. Recent researches and experiments show that clustering ensemble approaches can enhance the robustness and stabilities of unsupervised learning greatly. This paper makes an overview of the clustering ensemble approaches in recent years. It illustrates the contents and characteristics of recent clustering ensemble approaches research and discusses the future directions of clustering ensemble study.

关 键 词：聚类融合 数据重抽样 共识函数 差异度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]