一句话:博士提案研究用大语言模型识别与重构可维护性问题,并构建重构数据集、评测多种提示技术。
本博士提案研究如何利用大语言模型(LLM)进行代码重构,并报告三项研究。第一项研究评估 LLM 在识别与重构可维护性问题上的表现,结果表明尽管存在局限,但结果仍有前景。第二项研究提出 MaRV,一个面向提示技术设计的重构数据集。第三项为进行中的工作,计划评估多种提示技术在不同重构类型上的性能。提案还介绍了作者的贡献与博士阶段的未来议程。整体而言,该工作聚焦于将 LLM 应用于代码重构任务,并通过数据集构建与提示技术评测,系统探索其能力边界与改进方向。
This PhD proposal investigates the use of Large Language Models (LLMs) for code refactoring. We present and discuss three studies. The first evaluated LLMs in identifying and refactoring maintainability issues, showing that, despite their limitations, the results are promising. The second introduced MaRV, a refactoring dataset designed for prompting techniques. The third is an ongoing study in which we will evaluate the performance of several prompting techniques applied to different refactoring types. This proposal also presents our contributions and the future agenda of my PhD.