Research And Application Of SQL Injection Attack Classification Method Based On Deep Learning

At present,there have been many research results on the detection and defense of SQL injection attacks,but the classification of SQL injection attacks is not refined enough,and there is no application system that can automatically review SQL injection attack vulnerabilities.To this end,this paper is devoted to researching and constructing a refined classification model of SQL injection attacks,researching and applying deep learning algorithms to automatically classify SQL injection attacks,and designing and implementing a prototype system of SQL injection attack classification and vulnerability review.The main work of this paper:(1)Aiming at the shortcomings of the current classification methods of SQL injection attacks,based on the different characteristics and threat levels of different SQL attack types,a classification scheme of SQL injection attacks is proposed.(2)The research proposes a SQL injection attack feature modeling scheme that integrates Word2 Vec and TF-IDF.When the feature modeling obtains the sample sentence vector,the word frequency weight obtained by the TF-IDF algorithm is assigned to the word vector obtained by the Word2 Vec algorithm,and the sentence vector obtained by the sample is used together,so that the sample sentence vector contains both the contextual word semantic information and the keyword weight.It is beneficial to improve the accuracy of classification results.(3)Research and design the automatic classification model of CNN-SQL injection attack and LSTM-SQL injection attack based on semi-supervised learning.By optimizing the neural network structure and parameters,the automatic classification of SQL injection attack is realized.Verify the validity of the above research results.(4)Based on the above research results,the SQL injection attack classification and vulnerability automatic review system is designed and implemented,which verifies the availability of the research results in this paper.It is verified by experiments that the classification model trained based on the fusion feature modeling scheme proposed in this paper has better performance than the classification model trained using TF-IDF or Word2 Vec alone.CNNs perform better than LSTMs on classification models trained with the same feature modeling scheme.