用状态机控制推理流程，减少大模型检索时的无效思考。
From Token to Action: State Machine Reasoning to Mitigate Overthinking in Information Retrieval
- 设计离散动作（精炼、重排、停止）构成状态机，实现推理过程控制。
- 在BEIR/BRIGHT上提升性能3.4%，同时减少74.4%的令牌消耗。
- 无需调参即可适配不同模型与检索器，适合实际部署场景。
链式思维提示（CoT）虽能增强大语言模型在信息检索中的推理能力，但常导致过度思考，产生冗长且语义重复的推理轨迹，对结果无实质帮助。我们发现信息检索中存在两大问题：重复状态遍历和偏离用户意图的错误推理。为此，提出基于状态机的推理框架SMR，通过离散动作（精炼、重排、停止）实现早期终止与细粒度控制。在BEIR和BRIGHT基准上的实验表明，SMR在不需任务特定调优的情况下，将nDCG@10提升3.4%，同时减少74.4%的令牌使用。该方法可跨模型与检索器泛化，为传统CoT提供实用替代方案。代码与细节见https://github.com/ldilab/SMR。
原文摘要 · Abstract (English)
Chain-of-Thought (CoT) prompting enables complex reasoning in large language models (LLMs), including applications in information retrieval (IR). However, it often leads to overthinking, where models produce excessively long and semantically redundant traces with little or no benefit. We identify two key challenges in IR: redundant trajectories that revisit similar states and misguided reasoning that diverges from user intent. To address these, we propose State Machine Reasoning (SMR), a transition-based reasoning framework composed of discrete actions (Refine, Rerank, Stop) that support early stopping and fine-grained control. Experiments on the BEIR and BRIGHT benchmarks show that SMR improves retrieval performance (nDCG@10) by 3.4% while reducing token usage by 74.4%. It generalizes across LLMs and retrievers without requiring task-specific tuning, offering a practical alternative to conventional CoT reasoning. The code and details are available at https://github.com/ldilab/SMR.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。