SQL Injection Detection Technology Based on BiLSTM-Attention
Pengcheng Wen, Chengwan He, Wei Hua Xiong, Jihui Liu · 2021
SQL injection is one of the most popular and serious threats to information security. By exploiting the database vulnerability, the attacker may gain access to sensitive data, or enable the compromised computer to carry out further network attacks. Our research focuses on the application of neural network method to identify the injection characteristics of SQL injection string. The depth sequence model with loss layer (long-term and short-term memory and gating cycle unit) is adopted, and multi string analysis and word2vec are used for query string vectorization. By adding Attention mechanism to improve the BiLSTM model, a BiLSTM-Attention detection model is designed to train and test the data. The experimental results show that the accuracy of the SQL injection detection model based on BiLSTM is 99.3% and the recall is 98.2% on the actual data set. Compared with traditional machine learning method and neural network method, this method can identify SQL injection attack more efficiently.