中文摘要：

使用基于统计学习理论的支持向量机（Support Vector Machine，SVM）技术来构造垃圾邮件过滤系统．利用2个公开的邮件语料PUl和PU2来训练和测试过滤系统的性能．实验首先测试了语料的6种数据子集对过滤系统的分类错误率的影响情况，随后考察了采用不同类型核函数的SVMs准确率性能，最后考察了采用不同特征规模的数据集对过滤系统的影响．实验结果表明SVM技术是解决垃圾邮件过滤问题的一种很有效的方法．

英文摘要：

In this paper, we use support vector machine （SVM） as the spam filter. We study how classification error rate is affected when using different subsets of corpora, and explore the filters＇ accuracy when using SVMs with linear, polynomial, or RBF kernels. We also investigate the effect of the size of attribute set. Furthmore, we describe the architecture of the SVM-based anti-spam filter. Based on our experimental results and anlysis, we conclude that the SVM will be a very good alternative used to build anti-spam classifier, considering a good combination of accuracy, consistency, and speed.