Based On Deep Reinforcement Learning And Neural Network On XSS Attack Detecion Technology

Cross-site script(XSS)attacks are a common means of hackers around the web side development legacy vulnerabilities to launch attacks.The detection rate and accuracy of traditional XSS detection tools are limited to their attack corpora.XSS detection tools based on shallow machine learning cannot achieve good detection results and do not have generalization capabilities when facing large amounts of data.The adversarial attacks of artificial intelligence detection tools are also emerging,and the defense research against such attacks is also insufficient.In view of the shortcomings of the current XSS detection tools,which are limited by the design of high-quality detection models and the weak defense against adversarial attacks,an XSS detection model based on deep reinforcement learning and neural networks is proposed,which consists of an adversarial sample generator and an XSS detector.First of all,the adversarial attack defense against XSS detection tool is low,based on the adversarial learning idea of deep reinforcement learning and environment interaction,a model of XSS adversarial sample generator based on DQN algorithm is proposed.Its core idea is to mine XSS adversarial attack samples through the generator model for XSS detector to learn and train,so as to improve the detection model's adversarial attack defense capability.Then,the combined neural network of CNN and GRU is proposed,and its excellent feature extraction and learning ability is used to build the XSS detector model,so as to greatly improve the detection ability of XSS attack statements and make the overall model perform better.According to the above ideas,the model of adversarial sample generator based on DQN and the model of XSS detector based on CNN+GRU were designed and implemented respectively,and tens of thousands of positive and negative samples crawled and processed were used for the training and testing of three experiments.Experiment 1 is used for parameter selection of neural network.The results of experiment 2 show that compared with the XSS detection model based on CNN-GRU and the XSS detection model based on CNN and the XSS detection model based on CNN-LSTM,the XSS detection model based on CNN-GRU has stronger feature extraction and learning ability for XSS statements and better detection effect.In experiment 3,after eight rounds of cross-training,it was shown that the recognition rate of confrontational attack statements was greatly improved by the XSSdetector based on CNN+GRU through alternate training,and the adversarial sample generator based on the deep reinforcement learning algorithm DQN gradually increases the escape rate when generating the adversarial samples,the XSS detection system model as a whole has a certain XSS detection rate and defense ability against adversarial attacks.In summary,it is feasible and efficient to build XSS detection models using deep reinforcement learning and neural networks.