VulRep: vulnerability repair based on inducing commits and fixing commits

Abstract With the rapid development of information age, software vulnerabilities have threatened safety communication and mobile network, research on vulnerability repair is urgent. Different from existing machine learning-based approaches, we propose VulRep , a approach based introduction, which combines empirical findings inducing fixing commit with learning approaches for repair. Firstly, construct introduction dataset, generate AST tree code to form sequence after abstraction processing, input it into Transformer model recommendation list through beam search . After filling in abstracted code, combined rules defined by findings, final patch obtained verification. Experimental results show that can improve performance repairing vulnerabilities, illustrates effectiveness findings. In addition, found our more suitable type CWE-119 (Improper Restriction Operations within Bounds Memory Buffer) perform better.