cluster ensembles via weighted graph regularized nonnegative matrix factorization

ISCAS OpenIR

	cluster ensembles via weighted graph regularized nonnegative matrix factorization
	Du Liang; Li Xuan; Shen Yi-Dong
	2011
会议名称	7th International Conference on Advanced Data Mining and Applications, ADMA 2011
会议录名称	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
页码	215-228
会议日期	December 1
会议地点	Beijing, China
收录类别	EI
ISSN	0302-9743
ISBN	9783642258527
部门归属	(1) State Key Laboratory of Computer Science Institute of Software Chinese Academy of Sciences Beijing 100190 China; (2) Graduate University Chinese Academy of Sciences Beijing 100049 China
摘要	Cluster ensembles aim to generate a stable and robust consensus clustering by combining multiple different clustering results of a dataset. Multiple clusterings can be represented either by multiple co-association pairwise relations or cluster based features. Traditional clustering ensemble algorithms learn the consensus clustering using either of the two representations, but not both. In this paper, we propose to integrate the two representations in a unified framework by means of weighted graph regularized nonnegative matrix factorization. Such integration makes the two representations complementary to each other and thus outperforms both of them in clustering accuracy and stability. Extensive experimental results on a number of datasets further demonstrate this. © 2011 Springer-Verlag.; Cluster ensembles aim to generate a stable and robust consensus clustering by combining multiple different clustering results of a dataset. Multiple clusterings can be represented either by multiple co-association pairwise relations or cluster based features. Traditional clustering ensemble algorithms learn the consensus clustering using either of the two representations, but not both. In this paper, we propose to integrate the two representations in a unified framework by means of weighted graph regularized nonnegative matrix factorization. Such integration makes the two representations complementary to each other and thus outperforms both of them in clustering accuracy and stability. Extensive experimental results on a number of datasets further demonstrate this. © 2011 Springer-Verlag.
关键词	Clustering Algorithms Data Mining Factorization Graphic Methods
主办者	IBM Research; China Samsung Telecom R and D Center; Tsinghua University
语种	英语
内容类型	会议论文
URI标识	http://ir.iscas.ac.cn/handle/311060/16261
专题	中国科学院软件研究所
推荐引用方式 GB/T 7714	Du Liang,Li Xuan,Shen Yi-Dong. cluster ensembles via weighted graph regularized nonnegative matrix factorization[C],2011:215-228.