arXiv：cs.CL· Songze Li, Zhiqiang Liu, Zhaoyan Gong, Xiaoke Guo, Zhongpu Bo, Zhengke Gui, Lei Liang, Huajun Chen, Wen Zhang·· 4 小时前AI 评分37
Logits-to-Logic：让 LLM 在结构化知识推理中保持逻辑一致
Last Layer Logits to Logic: Empowering LLMs with Logic-Consistent Structured Knowledge Reasoning
AI 导读
针对 LLM 在知识图谱问答（KGQA）等结构化知识推理中的 Logic Drift 问题，研究者提出 Logits-to-Logic 框架，从自回归生成的 logits 输出入手，通过 logits 增强与 logits 过滤两个核心模块修正输出的逻辑缺陷。该工作被 EMNLP 2026 接收，在多个 KGQA 基准上取得 SOTA 表现，并显著提升 LLM 的逻辑一致性。
正文
Abstract:Large Language Models (LLMs) achieve excellent performance in natural language reasoning tasks through pre-training on vast unstructured text, enabling them to understand the logic in natural language and generate logic-consistent responses. However, the representational differences between unstructured and structured knowledge make LLMs inherently struggle to maintain logic consistency, leading to \textit{Logic Drift} challenges in structured knowledge reasoning tasks such as Knowledge Graph Question Answering (KGQA). Existing methods address this limitation by designing complex workflows embedded in prompts to guide LLM reasoning. Nevertheless, these approaches only provide input-level guidance and fail to fundamentally address the \textit{Logic Drift} in LLM outputs. Additionally, their inflexible reasoning workflows cannot adapt to different tasks and knowledge graphs. To enhance LLMs' logic consistency in structured knowledge reasoning, we specifically target the logits output from the autoregressive generation process. We propose the \textit{Logits-to-Logic} framework, which incorporates logits strengthening and logits filtering as core modules to correct logical defects in LLM outputs. Extensive experiments show that our approach significantly improves LLMs' logic consistency in structured knowledge reasoning and achieves state-of-the-art performance on multiple KGQA benchmarks.
| Comments: | Accepted by EMNLP 2026 Main |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2511.07910 [cs.CL] |
| (or arXiv:2511.07910v3 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2511.07910 arXiv-issued DOI via DataCite |
Submission history
From: Songze Li [view email]
[v1]
Tue, 11 Nov 2025 07:08:27 UTC (3,446 KB)
[v2]
Sat, 30 May 2026 07:08:14 UTC (3,628 KB)
[v3]
Fri, 2 Oct 2026 06:21:19 UTC (3,628 KB)
来源：arXiv：cs.CL · arxiv.org