Research On Cross-site Script Attack Detection Model Based On Machine Learning

With the improvement of the level of Internet technology,the Internet has greatly improved the daily life of human beings,and people's reliance on it has continued to increase.At the same time,security incidents on the Internet have also emerged.Among them,the crosssite scripting vulnerability is one of the vulnerabilities that have the greatest impact on Web applications in the Internet in recent years;and related research on this vulnerability has always attracted the attention of researchers in the security field.This paper analyzes the principle of cross-site scripting vulnerabilities.On the basis of summarizing domestic and foreign research,this paper studies the attack detection methods of cross-site scripting vulnerabilities and proposes two detection methods.Aiming at the problems of detection methods based on traditional rules and dynamic tests in detecting unknown and inefficient XSS attack scripts,this paper proposes an attack detection model based on single classification support vector machine.First,through the normalization of the attack samples,a sufficiently rich attack sample is obtained;then the n-gram-based TFDIF method is used for word segmentation to obtain the feature vector of the sample,and finally the model training is performed based on the single classification support vector machine,through the parameters Set up an optimal single-class detection model.Compared with the SVM model,the detection model based on the single classification support vector machine is not very effective in improving the recall rate.Therefore,this paper proposes a cross-site script attack detection model ABLSTM based on the two-way long and short-term memory model to solve this text classification problem.Based on the study of bidirectional LSTM,the Bi-LSTM model is improved by citing the attention mechanism to increase the model 's focus on features with high XSS correlation,thereby optimizing the model,not only improving the accuracy of model detection,but Recall rate has also improved significantly.