基于成对约束的主动半监督聚类算法  

Active semi-supervised clustering algorithm based on pairwise constraint

作　　者：李轶然[1] 张春娜[2] 

机构地区：[1]辽宁科技大学应用技术学院,辽宁鞍山114051 [2]辽宁科技大学软件学院,辽宁鞍山114051

出　　处：《计算机工程与设计》2013年第8期2897-2902,共6页Computer Engineering and Design

摘　　要：为了解决半监督聚类先验知识少、聚类偏差大的问题,提出了基于成对约束的主动半监督聚类算法。引入主动学习算法,增加约束集的信息量以使聚类效果更好;利用该约束集建立投影矩阵映射数据到低维空间,便于计算并提高聚类效果。算法中提出闭包替代思想,试图简化样本空间,以期获得降低聚类偏差的可能。由于聚类算法的实施对象是低维数据,成对约束集信息量大,聚类的时间效率以及性能均可保证。实验结果表明,采用主动学习的半监督聚类算法聚类效果提升显著,高效合理。In order to solve such problems as less priori knowledge and large deviation of clustering in semi-supervised clustering,active semi-supervised clustering algorithm based on pairwise constraint technique is proposed.First,the active learning algorithm is presented and the amount of constraint set is added to make the cluster effect better.At the same time,the projection matrix mapping data is built to low-dimensional space by using the constraint set,which is easy to calculate and can improve the clustering effect.In addition,idea of closure alternate is proposed,to simplify the sample space,and to obtain the possibility of reducing the clustering deviation.As the object of clustering algorithm implementation is a low dimensional data,pairwise constraints set information quantity is large,the time efficiency of clustering and performance can be guaranteed.Experimental results show that using active learning,semi-supervised clustering algorithm clustering effect is increased significantly and it is efficient and reasonable.

关 键 词：半监督聚类 主动学习 成对约束 约束集 K-MEANS 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]