XGV-BERT: Leveraging Contextualized Language Model and Graph Neural Network for Efficient Software Vulnerability Detection

The paper presents XGV-BERT, a new framework for software vulnerability detection. It combines the pre-trained CodeBERT model and Graph Neural Network (GCN) to achieve superior accuracy in detecting software vulnerabilities. By jointly training these modules, the model leverages large-scale pre-training and transfer learning. The results show that XGV-BERT significantly outperforms existing methods such as VulDeePecker and SySeVR in terms of accuracy. It achieves an F1-score of 97.5% and 95.5% on the VulDeePecker and SySeVR datasets respectively.

Publication date: 28 Sep 2023
Project Page: Not Provided
Paper: https://arxiv.org/pdf/2309.14677