基于一类SVM的不良信息过滤算法改进  

Research and Improvement of Filter Algorithm of Malicious Information Based on One-class SVM

作　　者：丁霄云[1] 刘功申[1] 孟魁[1] 

机构地区：[1]上海交通大学信息安全工程学院,上海200240

出　　处：《计算机科学》2013年第11A期86-90,114,共6页Computer Science

基　　金：973计划(2013CB329603);国家自然科学基金项目(61272441;61171173)资助

摘　　要：互联网的高速发展使得通过网络传输的文件监控和过滤成为一个热门课题。使用传统的基于字符串匹配的算法显然无法满足呈几何爆炸级别的信息增长的监管需求。而使用SVM确实可以提高分类效率,但依然存在维数过大导致存储资源和计算能力浪费的现象。为了有效减少SVM的维数,提出通过使用特征简约对向量机的维数进行约束的一个一类SVM算法改进。实验表明:在选用相同数量的特征词的前提下,改进算法使得不良信息分类和过滤的正确率有明显提高。The research of monitoring and filtering of the files transporting through internet is getting hotter and hotter now. The traditional algorithm based on string-matched is not able to meet the need of the huge increase of information. Although SVM model can surely improve the efficiency of the classification, the problem that SVM＇s too large dimension will affect the speed of examine still exists. It also causes a waste of storage space and compute ability. One algorithm was raised by first reducing the dimension by some specific algorithm before classification. The analysis result shows that after the improvement,we can get a more accurate result.

关 键 词：特征简约 一类SVM算法 分类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]