约束与度量相结合的半监督聚类集成  

Semi-supervised clustering ensemble combining constraints and metric

作　　者：韦斯婷 李志欣[1,2,3] 张灿龙[1,2,3] 

机构地区：[1]广西师范大学广西多源信息挖掘与安全重点实验室,广西桂林541004 [2]广西师范大学计算机科学与信息工程学院,广西桂林541004 [3]广西信息科学实验中心,广西桂林541004

出　　处：《计算机工程与设计》2016年第9期2440-2446,2452,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(61165009;61262005;61363035;61365009);国家973重点基础研究发展计划基金项目(2012CB326403);广西自然科学基金项目(2013GXNSFAA019345;2014GXNSFAA118368);"八桂学者"工程专项基金项目

摘　　要：针对基于约束的和基于度量的两种半监督聚类方法,结合聚类集成展开研究,提出一种结合这两种方法的半监督聚类集成方法。在少量先验知识辅助下使用这两种方法对样本聚类,产生不同的基聚类,运用一致性函数将基聚类整合得到目标聚类。基于度量的半监督聚类引入新的度量函数,适用于一般数据集,且因图像像素有着空间信息的特殊性,综合考虑像素固有特征和邻域空间分布,其适用于图像数据。在两类数据集上的实验结果表明,该方法大幅改善了聚类效果。In allusion to two kinds of semi-supervised clustering algorithms called constraint-based and metric-based,a semi-supervised clustering ensemble framework was provided which took both constraint-based and metric-based methods into account by combining cluster ensemble.With a small amount of prior knowledge,different base clustering partitions were generated respectively by using constraint-based method and metric-based method.The target clusters were obtained by integrating these base clustering partitions into a consensus function.The metric-based semi-supervised clustering introduced a new metric function.Given the spatial particularity of image pixels,this metric measure considered the spatial distribution of surrounding pixels besides inherent features of pixels,so that it was more applicable to image data clustering.Two groups of verification were conducted on general data sets and image data sets.Experimental results demonstrate that the proposed method produces considerable improvement to clustering performance.

关 键 词：半监督聚类 一致性函数 成对约束 度量学习 图像数据聚类 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]