arXiv 2601.035972026-01-20cs.CLcs.AI

从链式到图式：面向通用领域LLM的自结构化推理

From Chains to Graphs: Self-Structured Reasoning for General-Domain LLMs

University of Tokyo（东京大学）；Texas A&M University（德克萨斯大学）；University of Cambridge（剑桥大学）；Yale University（耶鲁大学）；Xiaomi EV（小米电动汽车）；Henan University（河南大学）

Yingjian Chen, Haoran Liu, Yinhong Liu, Sherry T. Tong, Aosong Feng, Jinghui Lu, Juntao Zhang, Yusuke Iwasawa, Yutaka Matsuo, Irene Li

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

本文提出自图推理框架SGR，通过结构化图表示提升LLM在通用领域问答中的推理一致性与准确性。

摘要：

大型语言模型（LLMs）在开放领域问答中展现出强大的推理能力，但其推理过程通常是线性的且常常逻辑不一致。相比之下，现实世界中的推理需要整合多个前提并并行解决子问题。现有方法，如思维链（CoT），以线性文本形式表达推理，虽然可能看起来连贯，但往往导致结论不一致。近期的方法依赖于外部提供的图结构，并未探索LLM如何构建和使用自身的图结构推理，尤其是在开放领域问答中。为填补这一空白，我们首次探索了LLM在通用领域问答中的图结构推理。我们提出了自图推理（SGR），一种框架，使LLM能够在生成最终答案前，显式地将推理过程表示为结构化的图。我们进一步构建了一个图结构推理数据集，将多个候选推理图合并为精细化的图结构以供模型训练。在五个涵盖通用和专业领域的问答基准测试中，实验表明SGR在推理一致性方面表现一致，并且在基模型上实现了17.74%的提升。经过SGR微调的LLaMA-3.3-70B模型与GPT-4o表现相当，并超越了Claude-3.5-Haiku，证明了图结构推理的有效性。

英文摘要：

Large Language Models (LLMs) show strong reasoning ability in open-domain question answering, yet their reasoning processes are typically linear and often logically inconsistent. In contrast, real-world reasoning requires integrating multiple premises and solving subproblems in parallel. Existing methods, such as Chain-of-Thought (CoT), express reasoning in a linear textual form, which may appear coherent but frequently leads to inconsistent conclusions. Recent approaches rely on externally provided graphs and do not explore how LLMs can construct and use their own graph-structured reasoning, particularly in open-domain QA. To fill this gap, we novelly explore graph-structured reasoning of LLMs in general-domain question answering. We propose Self-Graph Reasoning (SGR), a framework that enables LLMs to explicitly represent their reasoning process as a structured graph before producing the final answer. We further construct a graph-structured reasoning dataset that merges multiple candidate reasoning graphs into refined graph structures for model training. Experiments on five QA benchmarks across both general and specialized domains show that SGR consistently improves reasoning consistency and yields a 17.74% gain over the base model. The LLaMA-3.3-70B model fine-tuned with SGR performs comparably to GPT-4o and surpasses Claude-3.5-Haiku, demonstrating the effectiveness of graph-structured reasoning.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。