Vulnerability Code Similarity Detection Method Based on Transformer

With the expansion of software development, vulnerability detection and code reuse based on code similarity are becoming increasingly important. In the traditional twin network, there is a problem of insufficient feature extraction in the source code similarity evaluation. In this paper, a Transformer-based source code similarity evaluation model SCDetect is proposed, which effectively improves the performance of code similarity evaluation by using an improved self-attention mechanism (Si-AS), a Gaussian embedding method and a Transformer-based crossattention architecture. These improvements break through the limitations of traditional twin networks, improve feature extraction capabilities, and provide a new solution for code similarity evaluation.