Adaptive reasoning improves efficient and accurate text retrieval
Learning When to Think: Utility-Guided Adaptive Reasoning for Efficient Text Retrieval
Information RetrievalComputation and Language
Summary
Finding the right documents quickly in large collections is hard when you need to think through many steps. The authors created a system called AURA that learns when to keep thinking and when to stop, based on how well it tells apart relevant documents from confusing ones. This approach makes searches faster while keeping good accuracy, especially for difficult queries that need deep reasoning. It can also work well on more common search tasks without losing much quality.
What this means in practice
- •For search engine developers: Build search engines that adaptively decide how much reasoning to apply to queries, improving speed without losing accuracy on complex questions.
- •For digital library maintainers: Implement adaptive retrieval to speed up searching large text archives especially for in-depth research queries needing stepwise reasoning.
Authors
Yingtao Luo, Kangyi Zhao, Shiyi Du, Yubo Li
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.