Research On Cross-site Script Detection Method Based On Deep Convolutional Network

With the transformation and diversification of XSS attack methods,it make the detection more difficulty,and the harm is also increase.Due to the increasing number of malicious attack codes and the difficulty of identification,the traditional XSS attack detection model has been difficult to detect increasingly complex forms of attack.Therefore,this study applies the deep learning method to XSS attack detection.Deep learning can not only take advantage of learn more complex objective functions,but also obtain better features,and have greater advantage in deal with large sample size and high dimensional XSS data.For the traditional method,the XSS attack detection has the problems that the attack script less pertinence and the data detection is inaccurate,which is likely to cause low coverage of vulnerability detection.And the shallow machine learning methods are slow to train,and it is difficult to describe high-dimensional features.For the reason that,the XSS detection method of AMCNN-Highway deep network is proposed,which can accurately distinguish attacks and normal scripts from massive XSS data.The main work of this paper is as follows:(1)Firstly,200,000 XSS data for detection is collected and constructed.In order to enhance the recognizability of the data,the collected XSS data is processed by the Word2 vec method.The principle and realization process of convolutional neural network for natural language process are deeply studied,and analyze the TextCNN model,which is the most widely used deep learning model in natural language process;(2)In order to make the neural network pay attention to more representative malicious code features during the train process,the malicious code attention mechanism is introduced on the basis of the TextCNN structure.The input matrix is operated by the malicious code propensity algorithm to calculate the XSS corpus vector.Implemented an XSS detection model based on malicious code attention(ACNN).Experiments show that the method has higher detection accuracy than TextCNN;(3)Further deep optimization of the ACNN detection model to the AMCNN-Highway deep network.Deepen the number of layers of the neural network,and the multi-scale pooling operation is performed on the XSS feature map of the last three layers of the network in the AMCNN to obtain multi-layer features.The obtained feature vectors of each layer are fused,which is used as the input of the Highway network layer.The output optimized by the local adjustment of the Highway network is used as the input of the Softmax classification layer to complete the final XSS detection task.The optimal parameters of AMCNN-Highway neural network are obtained through experiments,and compared with other XSS detection methods,the accuracy of this model is better than other detection methods.The AMCNN-Highway neural network XSS detection method mainly reflects the advantages of four levels: it is beneficial to the analysis of high-dimensional feature data;the train speed of the network is faster;the generalization ability of the neural network is improved;improve the accuracy of the network for XSS detection,effectively avoid false negatives and false positives.