{"title":"An Improved Non-negative Matrix Factorization Algorithm for Combining Multiple Clusterings","authors":"Wei Wang","doi":"10.1109/MVHI.2010.72","DOIUrl":null,"url":null,"abstract":"Cluster ensemble has recently become a hotspot in machine learning communities. The key problem in cluster ensemble is how to combine multiple clusterings to yield a final superior result. In this paper, an Improved Non-negative Matrix Factorization (INMF) algorithm is proposed. Firstly, K-Means algorithm is performed to partition the hypergraph’s adjacent matrix and get the indicator matrix, which is then provided to NMF as initial factor matrix. Secondly, NMF is performed to get the basis matrix and coefficient matrix. Finally, clustering result is obtained via the elements in coefficient matrix. Experiments on several real-world datasets show that: (a) INMF outperforms the NMF-based cluster ensemble algorithm; (b) INMF obtains better clustering results than other common cluster ensemble algorithms.","PeriodicalId":34860,"journal":{"name":"HumanMachine Communication Journal","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2010-04-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"5","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"HumanMachine Communication Journal","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/MVHI.2010.72","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"Social Sciences","Score":null,"Total":0}
引用次数: 5
Abstract
Cluster ensemble has recently become a hotspot in machine learning communities. The key problem in cluster ensemble is how to combine multiple clusterings to yield a final superior result. In this paper, an Improved Non-negative Matrix Factorization (INMF) algorithm is proposed. Firstly, K-Means algorithm is performed to partition the hypergraph’s adjacent matrix and get the indicator matrix, which is then provided to NMF as initial factor matrix. Secondly, NMF is performed to get the basis matrix and coefficient matrix. Finally, clustering result is obtained via the elements in coefficient matrix. Experiments on several real-world datasets show that: (a) INMF outperforms the NMF-based cluster ensemble algorithm; (b) INMF obtains better clustering results than other common cluster ensemble algorithms.