基于马氏椭球学习机的多类文本分类算法  

Multiclass Text Classification Algorithm Based on Mahalanobis Hyper Ellipsoidal Learning Machine

作　　者：王祎[1] 秦玉平[2] 

机构地区：[1]渤海大学数理学院,辽宁锦州121013 [2]渤海大学工学院,辽宁锦州121013

出　　处：《渤海大学学报（自然科学版）》2014年第1期39-42,70,共5页Journal of Bohai University:Natural Science Edition

基　　金：辽宁省自然科学基金项目(No:201202003);辽宁省教育厅重点实验室项目(No:LS2010180)

摘　　要：提出一种基于马氏超椭球学习机的多类文本分类算法.对每一类训练样本,训练马氏超椭球学习机,使其包含该类尽可能多的样本,同时将噪音点排除在外.对于待分类样本,通过待分类样本的映射到每个超椭球球心的马氏距离确定其类别.实验结果表明,该算法提高了分类精度和分类速度.A new multiclass text classification algorithm based on mahalanobis hyper ellipsoidal learning ma -chine is proposed .To each class sample , training the hyper ellipsoidal learning machine , which include as much the class samples as possible and push the outlier samples away .For the sample to be classified , the distance from the sample mapping to the center of each mahalanobis hyper ellipsoidal are used to decide the sample classs.The results of the experiment show that the proposed algorithm has a higher classification accuracy and classification speed .

关 键 词：超椭球 噪音点 协方差矩阵 马氏距离 

分 类 号：TM315[电气工程—电机]