[论文解读] Atomizer: An LLM-based Collaborative Multi-Agent Framework for Intent-Driven Commit Untangling

Kangchen Zhu, Zhiliang Tian+3 more|arXiv (Cornell University)|Jan 3, 2026

Software Engineering Research被引用 0

一句话总结

简要结论：Atomizer 引入了一个多代理框架，使用意图导向的链式推理提示（Intent-Oriented CoT）和分组与评审者 Refinement 循环来解开复合提交，在准确性方面超越最先进的基于图的方法，尤其在复杂提交上表现更好。

ABSTRACT

Composite commits, which entangle multiple unrelated concerns, are prevalent in software development and significantly hinder program comprehension and maintenance. Existing automated untangling methods, particularly state-of-the-art graph clustering-based approaches, are fundamentally limited by two issues. (1) They over-rely on structural information, failing to grasp the crucial semantic intent behind changes, and (2) they operate as ``single-pass'' algorithms, lacking a mechanism for the critical reflection and refinement inherent in human review processes. To overcome these challenges, we introduce Atomizer, a novel collaborative multi-agent framework for composite commit untangling. To address the semantic deficit, Atomizer employs an Intent-Oriented Chain-of-Thought (IO-CoT) strategy, which prompts large language models (LLMs) to infer the intent of each code change according to both the structure and the semantic information of code. To overcome the limitations of ``single-pass'' grouping, we employ two agents to establish a grouper-reviewer collaborative refinement loop, which mirrors human review practices by iteratively refining groupings until all changes in a cluster share the same underlying semantic intent. Extensive experiments on two benchmark C# and Java datasets demonstrate that Atomizer significantly outperforms several representative baselines. On average, it surpasses the state-of-the-art graph-based methods by over 6.0% on the C# dataset and 5.5% on the Java dataset. This superiority is particularly pronounced on complex commits, where Atomizer's performance advantage widens to over 16%.

研究动机与目标

• Motivate the problem of composite commits and their maintenance burdens in software repositories.
• Address two key limitations of existing untangling methods: semantic understanding and lack of iterative refinement.
• Propose a multi-agent framework (Purifier, Profiler, Grouper, Reviewer) leveraging IO-CoT and refinement loops.
• Demonstrate empirical gains over baselines on benchmark C# and Java datasets.

提出的方法

• Purifier 构建最小变更子图（MCSs）以捕捉关键结构性变化。
• Profiler 应用意图导向链式推理（IO-CoT）来推断每个 MCS 的语义意图。
• Grouper 执行以意图为驱动的贪婪分组；Reviewer 提供反馈以在循环中 refined 分组。
• Iterative grouper-reviewer refinement loop 收敛到语义上连贯的原子提交。
• 评估将 Atomizer 与最先进的基线在 C# 和 Java 数据集上进行比较，报告变更节点的准确性。

实验结果

研究问题

• RQ1 如何在自动提交解耦中超越结构信号，有效融入开发者意图？
• RQ2 迭代的 grouper-reviewer refinement 循环是否优于单次遍历方法在分组质量上的提升？
• RQ3 基于 IO-CoT 的语义推理对复合提交的解耦准确性有何影响？
• RQ4 在基准的 C# 和 Java 数据集（包括复杂提交）上，Atomizer 相对于 SOTA 基于图的方法的性能如何？

主要发现

• Atomizer 实现了平均变更节点预测准确性 57%。
• Atomizer 在 C# 数据集上显著超越最先进基线（提升 ≥6.0%），在 Java 数据集上提升 ≥5.5%。
• 对于复杂提交，性能优势尤为显著，提升幅度超过 16%。
• 该框架整合了 IO-CoT 推理与 Grouper-Reviewer 循环，以缓解语义缺陷并实现迭代 refined。
• 在两个广泛使用的基准数据集上的实验显示出对现有基于图的方法的稳健改进。