基于图正则化MNMF的中文垃圾邮件过滤  

Chinese spam filtering based on graph regularized MNMF algorithm

作　　者：刘遵雄[1] 黄志强[1] 郑淑娟[2] 石菲[1] 

机构地区：[1]华东交通大学信息工程学院,南昌330013 [2]江西财经大学科研处,南昌330013

出　　处：《计算机应用研究》2013年第9期2672-2676,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(61065003);国家教育部人文社会科学研究规划基金资助项目(10YJC630379)

摘　　要：利用向量空间模型表示的文本邮件数据具有高维性,不利于邮件过滤模型的建立,需要对数据进行降维处理。最大间隔Semi-NMF(max-margin semi-nonnegative matrix factorization,MNMF)能够同时实现维数约减和邮件分类,而图正则化NMF能保持数据空间的几何结构。基于以上两种NMF改进模型,提出了图正则化MNMF(graph regularized MNMF,GMNMF)算法,并设计了一个迭代的求解算法。将GMNMF算法及其他相关算法用于中文垃圾邮件过滤实验,结果表明GMNMF算法构建的过滤模型要优于其他较好的算法构建的过滤模型。Text e-mail data represented by vector space model （VSM） are high dimensionality. This situation is not conducive to construct e-mail filtering model. Therefore, dimensionality reduction need be performed. MNMF could simultaneously achieve dimensionality reduction and e-mail classification, and graph regularized NMF could keep the geometrical structure of the data space. Based on the above two improved NMF models, this paper put forward GMNMF algorithm, and designed an iterative solution algorithm. Using GMNMF algorithm and other related algorithm do Chinese spam filtering experiments. The experimental results show that the model of the proposed algorithm is superior to models of other good algorithms.

关 键 词：向量空间模型 维数约减 最大间隔Semi—NMF 图正则化MNMF 中文垃圾邮件过滤 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]