大数据中基于熵加权的稀疏分数特征选择聚类算法  

Clustering algorithm based on entropy weighted and sparse fractional feature selection in big data

作　　者：魏霖静[1] 宁璐璐[2] 郭斌 侯振兴[4] Wei Linjing;Ning Lulu;Guo Bin;Hou Zhenxing(School of Information Science ＆ Technology,Gansu Agriculture University,Lanzhou 730070,China;School of Biological Sciences,Nanyang Technological University,Singapore City 639798,Singapore;School of Computer Science ＆ Engineering,Hohai University,Nanjing 210094,China;School of Information Science ＆ Engineering,Nanjing University,Nanjing 210093,China)

机构地区：[1]甘肃农业大学信息科学技术学院,兰州730070 [2]南洋理工大学生物科学学院,新加坡639798 [3]河海大学计算机与信息学院,南京210094 [4]南京大学信息管理学院,南京210093

出　　处：《计算机应用研究》2018年第8期2293-2294,2303,共3页Application Research of Computers

基　　金：国家自然科学基金资助项目(61063028;31560378);江苏省自然科学基金青年基金资助项目(BK20150784);中国博士后面上资助项目(2015M581800);甘肃省科技支撑计划项目(1604WKCA011);陇原青年创新创业人才项目(2016-47);2016年度甘肃省高校重大软科学(战略)研究项目计划资助项目(2016F-10)

摘　　要：为了提高大数据统计及分析的效率,有必要对数据集合进行聚类,以减少数据集合维度,并去掉相似数据冗余。采用熵加权和稀疏分数特征选择相结合,一方面对异构数据进行局部结构划分,降低数据维度,对局部结构的特征重要性标记并排序,提高聚类精度,另一方面,提高聚类稳定性。实验证明,该方法对不同种类的大数据聚类具有较强的适用性。In order to improve the efficiency of data statistics and analysis, it is necessary to cluster data sets, for reduces the data sets collection dimension and removes similar data redundancy. This paper used entropy weighted and sparse fractional feature selection. On the one hand, it divided the local structure of heterogeneous data, reduced the data dimension, marked and sorted the feature importance of local structure, and improved the clustering accuracy. Experimental results show that the method has strong applicability to different kinds of large data clustering.

关 键 词：数据聚类 熵加权 稀疏分数 特征选择 数据维度 大数据 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]