基于奇异值分解的自适应近邻传播聚类算法  

Self-adaptive Affinity Propagation Clustering Algorithm Based on Singular Value Decomposition

作　　者：王丽敏[1] 姬强[1] 韩旭明[2] 黄娜[1,3] 

机构地区：[1]吉林财经大学管理科学与信息工程学院,长春130117 [2]长春工业大学软件学院,长春130012 [3]上海财经大学信息管理与工程学院,上海200433

出　　处：《吉林大学学报（理学版）》2014年第4期753-757,共5页Journal of Jilin University:Science Edition

基　　金：国家自然科学基金(批准号:61202306);吉林省科技厅项目(批准号:20100507;201215119;20130522177JH);吉林省教育厅重点规划项目(批准号:2012185);吉林省高校新世纪优秀人才支持计划项目(批准号:2014159);吉林财经大学青年学俊支持计划项目

摘　　要：针对近邻传播算法无法有效处理高维数据而导致聚类效果不佳的问题,提出一种基于奇异值分解的自适应近邻传播(SVD-SAP)聚类算法.通过引入奇异值分解,对高维数据进行重构、降维,消除冗余信息,并在此基础上采用非线性函数策略,自适应地调整阻尼系数,提高算法的聚类性能.仿真实验结果表明,与已有算法相比,该改进算法聚类精度更高,收敛速度更快.Aiming at the problem that affinity propagation algorithm has a difficult to deal with high- dimensional data, the authors put forward an self-adaptive affinity propagation algorithm based on singular value decomposition. With the aid of singular value decomposition introdued, the high dimensional data were reconstructed and dimensions were reduced to eliminate the redundant information, based on which, a nonlinear function strategy was adopted to adjust the damping factor adaptively and improve the clustering performance of the algorithm. Experimental results show that the proposed algorithm has obviously better clustering performance than the traditional algorithm on clustering accuracy and the number of iterations.

关 键 词：近邻传播聚类算法 奇异值分解 非线性函数策略 阻尼系数 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]