基于支持向量机的兼类文本分类算法研究  

Study on multi-subject text classification algorithm based on support vector machines

作　　者：秦玉平[1] 艾青[1] 王秀坤[1] 李祥纳[1] 刘卫江[1] 

机构地区：[1]大连理工大学电子与信息工程学院,渤海大学信息科学与工程学院,东南大学计算机科学与技术学科博士后流动站,辽宁大连116024,辽宁锦州121000,江苏南京210096

出　　处：《计算机工程与设计》2008年第2期408-410,共3页Computer Engineering and Design

基　　金：国家973重点基础研究发展计划基金项目(2001CCA00700);国家自然科学基金项目(90104031)

摘　　要：针对兼类文本,提出了两种基于支持向量的分类算法。一种是采用1-a-1方法训练子分类器,通过子分类器得到待分类样本的隶属度矩阵,依据隶属度矩阵每行元素和判定该文本所属类别。另一种是采用1-a-r方法训练子分类器,通过子分类器得到待分类样本的隶属度向量,根据隶属度向量判定该文本所属的类别。实验结果表明,这两种算法都具有较好的准确率、召回率和F1值。For multi-subject text, two classification algorithms based on support vector machines are proposed. The first method uses 1-a-1 to train sub-classifiers, for the samples to be classified, sub-classifiers are used to obtain membership matrix, and then according to the sum of every line of membership matrix, confirms the subjects that the sample belongs to. The second method uses 1-a-r to train sub-classifiers, for the samples to be classified, sub-classifiers are used to obtain the membership vector, according to the membership vector, confirms the subjects that the sample belongs to. The experimental results show that the proposed algorithms have higher performance on precision, recall and F 1 value.

关 键 词：支持向量机 隶属度矩阵 隶属度向量 召回率 准确率 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]