一句话：提出多智能体协作的 LLM 代码审查者推荐方法 CoRe，在四个项目上超越现有方法并提升可解释性。
将 pull request 分配给合适的代码审查者能够加速审查流程并帮助发现潜在缺陷，但 pull request 与审查者本身的复杂性使得二者之间的匹配十分困难。已有研究多从 pull request 的语义信息或审查者的画像信息入手，却忽略了二者之间由多种因素和策略共同刻画的内在关联，因而推荐准确率不理想。为此，本文提出 CoRe——一种强调灵活性与适应性的协作式智能体代码审查者推荐方法。该方法利用大语言模型精确捕捉 pull request 与审查者双方的丰富文本语义，并借助多智能体系统强大的规划、协作与决策能力，将多种因素整合进推荐过程，从而显著提升基于 LLM 的代码审查者推荐效果。作者在四个被广泛使用的项目上进行了评估，结果表明 CoRe 在性能与可解释性两方面均优于当前最先进的方法。
Assigning pull requests to appropriate code reviewers can accelerate the review process and help uncover potential bugs. However, the inherent complexities in pull requests and code reviewers present challenges in making suitable matches between them. Prior studies focus on mining rich semantic information from pull requests or profile information from code reviewers to improve efficiency. These approaches often overlook the intrinsic relationships between pull requests and code reviewers, which can be represented by a combination of multiple factors and strategies, resulting in suboptimal recommendation accuracy.To address this issue, we propose CoRe, a collaborative agent-based code reviewer recommendation approach that emphasizes flexibility and adaptability. We leverage Large Language Models (LLMs) to precisely capture the rich textual semantics of both pull requests and reviewers. Additionally, we integrate various factors into the recommendation process through the robust planning, collaboration, and decision-making capabilities of multi-agent systems. This integration significantly enhances the performance of LLM-based code reviewer recommendations. We evaluate the effectiveness of our approach on four widely used projects. The results demonstrate that CoRe outperforms state-of-the-art methods in both performance and interpretability.CCS Concepts• Software and its engineering → Collaboration in software development.
@inproceedings{WangZZLCZ024,
title = {Unity Is Strength: Collaborative LLM-Based Agents for Code Reviewer Recommendation},
author = {Luqiao Wang and Yangtao Zhou and Huiying Zhuang and Qingshan Li and Di Cui and Yutong Zhao and Lu Wang},
booktitle = {ASE},
year = {2024},
doi = {10.1145/3691620.3695291}
}