DCFA: Dual-view Causal-inspired Attribution for Failure Reasoning in LLM-based Multi-agent Systems

📄 arXiv: 2609.04749v1 📥 PDF

作者: Zehao Wang, Lanjun Wang, Shilong Jin, Junjie Chen, Yanghua Xiao

分类: cs.AI

发布日期: 2026-09-04


💡 一句话要点

提出DCFA以解决LLM多智能体系统中的故障归因问题

🎯 匹配领域: 支柱九:具身大模型 (Embodied Foundation Models)

关键词: 故障归因 多智能体系统 大型语言模型 因果推理 反事实推理 系统鲁棒性 推理错误

📋 核心要点

  1. 现有方法在故障归因中常常只捕捉到小的偏差,未能识别系统故障的决定性原因,导致归因效果不佳。
  2. DCFA通过构建因果依赖图和应用局部反事实推理,提出了一种无训练的故障归因框架,有效识别初始决定性错误。
  3. 在Who&When基准测试中,DCFA在六种LLM上实现了最高8.27%的逐步准确率提升,显示出其有效性。

📝 摘要(中文)

基于大型语言模型(LLM)的多智能体系统近年来快速发展,然而这些系统仍然脆弱,常常出现推理和协调错误,导致系统级故障。故障归因依赖于追踪智能体之间的自然语言交互,以识别决定性错误,即修正后能逆转系统故障的最早行动。现有方法面临两个主要挑战:一是浅层归因,往往只捕捉到小的偏差,未能识别系统故障的决定性原因;二是上下文退化,随着系统追踪长度增加,模型的推理能力迅速下降。为此,本文提出了DCFA,一个无训练的故障归因框架,通过构建结构化的因果依赖图和局部反事实推理来识别初始决定性错误。实验结果表明,DCFA在Who&When基准上相较于最先进的基线提高了最高8.27%的逐步准确率。

🔬 方法详解

问题定义:本文旨在解决LLM多智能体系统中的故障归因问题,现有方法往往只能捕捉到表面的小错误,未能深入识别导致系统故障的根本原因。

核心思路:DCFA的核心思路是通过构建因果依赖图和局部反事实推理,系统性地识别出初始的决定性错误,从而提升故障归因的准确性。

技术框架:DCFA框架包括两个主要模块:全局模块负责从系统追踪中构建结构化的因果依赖图,局部模块则应用反事实推理来细化归因结果。

关键创新:DCFA的创新在于其无训练的设计理念,结合了因果推理和反事实推理,能够有效克服现有方法在故障归因中的局限性。

关键设计:在设计中,DCFA采用了结构化的因果图构建方法,确保了信息的有效传递,并通过局部推理机制增强了模型的推理能力。

🖼️ 关键图片

fig_0
fig_1
fig_2

📊 实验亮点

实验结果显示,DCFA在Who&When基准测试中,相较于最先进的基线模型,逐步准确率提升了最高8.27%,证明了其在故障归因中的有效性和优势。

🎯 应用场景

该研究的潜在应用领域包括智能客服、自动驾驶、机器人协作等多智能体系统,能够有效提高系统的鲁棒性和故障处理能力,具有重要的实际价值和未来影响。

📄 摘要(原文)

Large language model (LLM)-based multi-agent systems have experienced rapid growth in recent years. Despite their promise, such systems remain fragile, frequently exhibiting reasoning and coordination errors that can lead to system-level failures. Failure attribution in such systems relies on tracing natural language interactions among agents to identify the decisive error, which refers to the earliest action whose correction can reverse system failure. There are two key challenges: 1) Shallow attribution: Existing methods often capture only minor deviations, such as incomplete retrievals or formatting errors, which verification mechanisms can correct, while missing the decisive cause of system failure. 2) Contextual degradation: As the length of the system traces increases, the model's reasoning ability rapidly deteriorates. To address these challenges, we propose DCFA, a training-free framework for failure attribution. DCFA integrates a global module that constructs structured causal-inspired dependency graphs from system traces to identify the initial decisive error, and a local module that applies local counterfactual-inspired reasoning to refine causal-inspired attribution. Experiments on the Who&When benchmark across six LLMs show that DCFA improves step-level accuracy by up to 8.27% over state-of-the-art baselines.