Vulnerability Detection with Deep Learning on a Multilingual Software

The structure of existing open source software is becoming more and more complex, and the development of such software is often written in multiple programming languages, and security vulnerabilities caused by different types of programming languages are more common in the actual environment. The existing vulnerability detection technology mainly focuses on feature learning for a single programming language, which can not achieve high-precision vulnerability detection for different programming language projects. So we propose that Vulnerability Detection with Deep Learning on a Multilingual software (VuDeMu). VuDeMu constructs a deep learning model based on Bidirectional Gate Recurrent Unit-Conditional Random Field(BGRU-CRF), which is used for Named Entity Recognition(NER) of functions in various programming languages in multilingual software, and reconstructs it with the program slice results to reduce the loss of grammatical and semantic information in the process of code representation, finally, VuDeMu designs the Bi-directional Long Short-Term Memory (BLSTM) to extract features of vulnerability code to achieve the effect of vulnerability detection in multilingual software. The comprehensive experimental results on the NVD show that the detection accuracy of this method to identify software vulnerabilities on the multilingual software vulnerability datasets reaches 93.5%, and the F1 reaches 92.3%. Compared with the latest VulDeePecker, SySeVR, AE-KNN methods, the detection accuracy of VuDeMu is 3.06% higher than the average, and the Recall is 3.87% higher than the average.