基于提示工程的静态调用图剪枝方法OA

Existing pruning methods face challenges such as low quality of model datasets,high training costs,insufficient understanding of code semantics,and difficulty in handling tasks with long chains of reasoning.To address these issues,we propose a static call graph pruning method called PCGPruner based on prompt engineering.PCGPruner utilized prompts to guide large language models in conjunction with code structure information and semantic details to perform call relationship inference tasks.Experimental results demonstrate that PCGPruner performs exceptionally well in static call graph pruning tasks,with an F1-score of 0.87,which is 30.0%higher than the baseline,and PCGPruner can effectively improve the quality of call graphs.