Research On XSS Detection Based On Bayesian Network

Despite only a few decades of history,the Internet has become one of the fastest growing and most vigorous industries in the world.From the advancement of Web 1.0 to the Web 2.0 era,the Internet application model has become more complicated while providing more convenient and efficient services.The security vulnerabilities exposed in application services have also increased dramatically.The losses are also huge.Among the many network security threats,cross-site scripting attacks are a form of attack that has a long time span,a wide coverage,and a large potential damage loss.Effective detection of this type of attack is of great significance for purifying the Internet environment and protecting privacy and property security.In order to promote the research of cross-site scripting attack detection,this paper proposes a cross-site scripting attack detection method BDXSS based on Bayesian network.Research was conducted on the domain knowledge of cross-site scripting attacks and Bayesian network construction methods to solve the following two problems: One is the construction of indicator models,the purpose is to form a comprehensive and effective indicator model in the field of cross-site scripting attacks to improve detection performance;Second,for the realization and evaluation of the method in this paper,we need to select evaluation indicators to test the method in this paper.The research work of this paper mainly includes the following three parts:(1)An indicator model for cross-site scripting attack detection is constructed.By studying the knowledge of cross-site scripting attack domains,and based on the study of attack payload samples,an indicator model is constructed,including three sub-categories based on events,tag attributes,and experience based on a total of 54 dimensions.(2)The detection method BDXSS for cross-site scripting attack is proposed.This method combines an indicator model constructed based on domain knowledge and a Bayesian network detection method.In terms of model building methods of Bayesian networks,feasible learning strategy for model training is selected,which is summarized as a model training algorithm based on BIC scoring function,HC search strategy and Bayesian model estimation.(3)Experiments verify the effectiveness of this method.The experiment is designed under the guidance of the detection method to test the detection performance of the detection method proposed in this paper.Controlled trials are designed,and the detection effect of Bayesian network is compared and evaluated through three sets of controlled trials.Under the condition that the training and test sets are randomly divided and the average value is repeated five times,the detection effect achieved by the BDXSS detection method proposed in this paper is: 0.99539 for accuracy rate,0.99270 for recall rate measure,0.99689 for precision rate measure and 0.99479 for F1 value measure.In the three evaluation indicators of accuracy rate,recall value and F1 value,it exceeds the level of all control groups,and for one group of control groups,all four evaluation indicators are leading.