一种基于多模态特征融合的垃圾邮件过滤方法

A Spam Filtering Method Based on Multi-modal Feature Fusion

近年来,垃圾邮件制造者为了逃避基于文本的垃圾邮件过滤系统的检测,将垃圾信息嵌入到图像中,并将其附着在邮件正文中进行传播。传统的基于文本的过滤方式无法处理此类包含垃圾信息的邮件图像。为了应对这种同时包含文本和图像的垃圾邮件,本文提出了一种基于多模态特征的融合文本、图像等多媒体信息的过滤方法。首先通过抽取邮件的文本特征和图像特征构建多个分类器,然后采用多分类器融合技术对各分类器的输出结果进行综合。通过对TREC垃圾邮件语料集的测试实验表明,本文提出多模态特征融合的方法获得了比单个分类器更好的效果,准确率达到90%以上。

In recent years,for escaping the text-based spam filtering detection system,the spammers insert junk information into the image and attach it to the message body.In order to deal with such spam that contains text and images,a new filtering method is proposed,which fuses text,image and other multi-modal features by extracting the text features and image features to build multiple classifiers,and by employing multiple classifier fusion technology to integrate the output of each classifier.The experimental result on TREC dataset show that the fusion method achieves a better result than that of a single classifier and can achieve over 90% in accuracy rate.