arXiv cs.CL· Yingtao Luo, Kangyi Zhao, Shiyi Du, Yubo Li·· 2 天前AI 评分22
AURA：面向高效文本检索的效用引导自适应推理框架
Learning When to Think: Utility-Guided Adaptive Reasoning for Efficient Text Retrieval
AI 导读
AURA 框架通过直接监督中间推理状态对相关文档与困难干扰项的区分能力，并训练轻量效用估计器在推理时自适应调整推理深度。在 BRIGHT 基准上，AURA-4B 以平均 2.8 个推理步骤和每查询 96 个生成 token 达到 34.7 nDCG@10，自适应停止将延迟从 4.6 秒降至 0.76 秒。MS MARCO 上的结果表明，这种专门化不会显著降低常规域内检索性能。
正文 · 原文
Abstract:Reasoning-conditioned embeddings can improve retrieval by progressively refining semantic constraints. However, additional reasoning may introduce over-generalized or semantically noisy abstractions that weaken retrieval specificity. Existing methods largely supervise reasoning through final retrieval outcomes, providing limited guidance over the refinement process itself. We propose AURA, a framework that directly supervises intermediate reasoning states according to how well they separate relevant documents from hard distractors. AURA further trains a lightweight utility estimator over prefix-conditioned embeddings to adapt reasoning depth at inference time. On the reasoning-intensive BRIGHT benchmark, AURA-4B achieves 34.7 average nDCG@10 while using 2.8 reasoning steps and 96 generated tokens per query. Under matched deployment conditions, adaptive stopping reduces latency from 4.6 to 0.76 seconds relative to full-trajectory reasoning. Results on MS MARCO further suggest that this specialization need not substantially degrade conventional in-domain retrieval.
| Comments: | Accepted to EMNLP 2026 Findings |
| Subjects: | Information Retrieval (cs.IR); Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.21030 [cs.IR] |
| (or arXiv:2610.21030v1 [cs.IR] for this version) | |
| https://doi.org/10.48550/arXiv.2610.21030 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Yingtao Luo [view email]
[v1]
Thu, 15 Oct 2026 00:04:21 UTC (323 KB)
来源：arXiv cs.CL · arxiv.org