让检索模型像人一样思考，逐步推理出答案。
Integrating Chain-of-Thought into Generative Retrieval: A Preliminary Study
- 在生成文档编号时穿插思维链，实现边思考边检索。
- 在4个复杂检索数据集上平均提升6.86%，达到当前最优。
- 适合需要多步推理的智能问答与深度检索任务。
尽管生成式检索（GR）在标准检索基准上表现良好，但现有方法直接将查询映射为文档标识符（docids），缺乏中间推理过程，限制了其在需要多步推理的复杂查询上的效果。作为将思维链（CoT）融入生成式检索的初步研究，我们提出ThinkGR，一个统一框架，将思维链与docid生成交织在一个生成过程中，支持迭代式思考与检索。为弥合自由形式思维生成与结构化检索目标之间的差距，我们设计了：(1) 动态切换无约束思维生成与受限docid解码的混合解码策略；(2) 两阶段训练方法：先通过监督微调对齐思维-检索模式，再通过基于检索结果的强化学习优化思维质量。在四个多跳检索基准上的实验表明，ThinkGR实现了最先进的性能，平均提升达+6.86%。本工作为增强生成式检索的显式推理能力开辟了新路径，对需复杂推理的检索任务具有重要启示。
原文摘要 · Abstract (English)
While generative retrieval (GR) demonstrates competitive performance on standard retrieval benchmarks, existing approaches directly map queries to document identifiers (docids) without intermediate deliberation, limiting their effectiveness for complex queries that require multi-step reasoning. As a preliminary study on integrating chain-of-thought (CoT) into generative retrieval, we introduce ThinkGR, a unified framework that interleaves CoT with docid generation, enabling iterative thinking and retrieval within a single generative process. To bridge the gap between free-form thought generation and structured retrieval targets, we design (1) a hybrid decoding strategy that dynamically switches between unconstrained thought generation and constrained docid decoding, and (2) a two-phase training approach that first aligns thought-retrieval patterns through supervised fine-tuning, then optimizes thought quality via retrieval-grounded reinforcement learning. Experiments on four multi-hop retrieval benchmarks demonstrate that ThinkGR achieves state-of-the-art performance with an average improvement of +6.86\%. Our work opens new avenues for enhancing generative retrieval with explicit deliberation capabilities, with promising implications for retrieval tasks requiring complex reasoning.
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。