双向聚类方法综述

传统的聚类方法由于无法提取样本和变量间的局部对应关系,并且当数据具有高维性和稀疏性时表现不佳,因此学者们提出了双向聚类,基于样本和变量间的局部关系,同时对样本和变量进行聚类,形成一系列子矩阵的聚类结果。近年来,双向聚类发展迅速,在基因分析、文本聚类、推荐系统等领域应用广泛。首先,对双向聚类方法进行梳理与归纳,重点阐述稀疏双向聚类、谱双向聚类和信息双向聚类三类方法,分析它们之间的区别和联系,并且介绍这三类方法在多源数据的整合分析、多层聚类、半监督学习以及集成学习上的发展现状和趋势;其次,重点介绍双向聚类在基因分析、文本聚类、推荐系统等领域的应用研究情况;最后,结合大数据时代的数据特征和双向聚类存在的问题,展望双向聚类未来的研究方向。

Review of Biclustering

The duality between samples and variables cannot be exploited by traditional clustering methods,which often performs badly when dealing with sparse and high-dimensional data.The scholars then proposed biclustering,which can cluster both samples and variables simultaneously to get submatrices,based on the duality between samples and variables.In recent years,biclustering methods have developed rapidly and been widely used in many areas,such as microarray analysis,text clustering,recommendation system and so on.This paper at first reviews biclustering methods and focuses on three classical biclustering methods,including Sparse Biclustering,Spectral Biclustering and Information Theoretic Coclustering.In detail,we conclude the differences and relationships between these three methods,and introduce the development status and trends of integrative analysis in multi-source datasets,multi-level clustering,semi-supervised learning,supervised learning and ensemble learning.Secondly,we focus on the application research of biclustering in the fields of microarray analysis,text clustering and recommendation system.Finally,combining the data characteristics of the big data era with the existing problems in biclustering,the future research direction of biclustering is discussed.