改进编码-解码框架下的跨站脚本检测

Cross-site script detection based on improved Encoder-Decoder framework

为解决传统机器学习方法特征提取工作艰难导致对跨站脚本检测性能有限的问题,提出应用注意力机制改进编码-解码框架的方法并以此建立模型检测跨站脚本。由卷积神经网络和双向门控循环单元网络并行构成编码器,既考虑输入数据上下文信息,又充分提取有效特征;使用注意力机制解决传统编码-解码框架的“分心问题”;使用门控循环单元网络构成解码器,使用分类器进行分类检测。在收集到的数据集上进行仿真实验,验证了模型的有效性和性能优势。

To solve the problem that feature extraction in traditional machine learning methods is difficult,which leads to the limited performance of cross-site script detection,a method to improve the Encoder-Decoder framework using attention mechanism was proposed,and a model for cross-site script detection was established.The Encoder was constructed using the convolutional neural network and the bidirectional gated recurrent unit network in parallel,which considered both the input data context information and the extraction of the effective features.The attention mechanism was used to solve the distraction problem of the traditional Encoder-Decoder model framework.The gated recurrent unit network was used to form the Decoder,and the classi-fier was used for classification detection.Simulation experiments were carried out on the collected data sets to verify the effectiveness and performance advantages of the model.