基于深度学习的跨站脚本攻击检测技术的研究  

Detection Technology Research of Cross-site Scripting Attack Based on Deep Learing

作　　者：吴金宇 陶文伟[1] 张富川 江泽铭 王依云 赵宇珩 王宝会[2] WU Jinyu;TAO Wenwei;ZHANG Fuchuan;JIANG Zeming;WANG Yiyun;ZHAO Yuheng;WANG Baohui(China Southern Power Grid Co.,LTD,Guangzhou,510623,China;College of Software,Beihang University,Beijing,100191,China)

机构地区：[1]中国南方电网有限责任公司,广州510623 [2]北京航空航天大学软件学院,北京100191

出　　处：《网络新媒体技术》2023年第5期33-44,共12页Network New Media Technology

摘　　要：针对互联网应用的网络攻击。跨站点脚本XSS攻击是常见的针对Web应用程序的攻击种类。本文提出的基于深度学习的XSS检测模型,将CNN神经网络和BiLSTM神经网络序列化,融合两者优点学习样本的局部特征和上下文依赖特征,并通过Attention注意力机制加权计算来解决长序列效果差的问题,并融合BERT预训练的特征向量加速模型训练,提高检测效果,从而实现XSS检测模型的构建。优化的模型能够自动的提取样本的特征信息并完成分类检测相对于传统的静态和动态检测方法及采用人工特征提取的机器学习算法相比,在准确率和误报率方面都有较大提升。准确率、召回率、精确度值超过目标值(98%),误报率低至0.12%。Cross site scripting(XSS)attacks are common types of attacks against Web applications.The XSS detection model based on deep learning proposed in this paper combines the advantages of CNN neural network and BiLSTM neural network,integrates the local features and context dependent features of the learning samples with their advantages,solves the problem of poor effect of long sequences through the weighted calculation of Attention mechanism,and accelerates the model training with the feature vector pre trained by BERT to improve the detection effect,so as to realize the construction of XSS detection model.The optimized model can automatically extract the feature information of samples and complete classification detection.Compared with traditional static and dynamic detection methods and machine learning algorithms using artificial feature extraction,it has a greater improvement in accuracy and false alarm rate.The accuracy rate,recall rate and precision value exceed the target value(98%),and the false alarm rate is as low as 0.12%.

关 键 词：跨站脚本攻击 深度学习 网络安全 检测技术 特征 

分 类 号：TP393.08[自动化与计算机技术—计算机应用技术] TP18[自动化与计算机技术—计算机科学与技术]