基于超球支持向量机的多主题文本分类算法  

Multi-subjects text classification algorithm based on hyper-sphere support vector machines

作　　者：艾青[1] 秦玉平[2,3] 李迎春[1] 

机构地区：[1]辽宁科技大学软件学院,辽宁鞍山114051 [2]大连理工大学电子与信息工程学院,辽宁大连116024 [3]渤海大学信息科学与工程学院,辽宁锦州121000

出　　处：《计算机工程与设计》2010年第10期2273-2275,2279,共4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);公安部基金项目(20032129001)

摘　　要：针对标准支持向量机多分类算法不能解决多主题文本分类问题,提出了一种基于超球支持向量机的多主题文本分类算法。该算法用超球支持向量机训练得到每个超球,计算待分类文本到每个超球球心的距离,依据距离得到隶属度向量,最后根据隶属度向量判定该文本所属的主题。实验结果表明,该算法具有更好的召回率、准确率和F1值。For the problem that multi-subjects text classification can not be solved by standard SVM multi-class classification algorithms,a multi-subjects text classification algorithm based on hyper-sphere support vector machine is proposed.Hyper-sphere support vector machine is used to obstain each hyper-sphere,the distances from the text to be classified to center of the every hyper-sphere is computed,according to the distances,the membership vector is gotten,at last label the subjects that the text belong to according to the membership vector.The experimental results show that the algorithm has higher performance on recall,precision,and F1.

关 键 词：超球支持向量机 隶属度向量 召回率 准确率 F1值 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]