Pith. sign in

REVIEW

SAGE: Strategy-Adaptive Generation Engine for Query Rewriting

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2506.19783 v2 pith:7ZPIEZFZ submitted 2025-06-24 cs.AI

SAGE: Strategy-Adaptive Generation Engine for Query Rewriting

classification cs.AI
keywords shapinggenerationretrievalrewardsagestrategiesconciseengine
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Query rewriting is pivotal for enhancing dense retrieval, yet current methods demand large-scale supervised data or suffer from inefficient reinforcement learning (RL) exploration. In this work, we first establish that guiding Large Language Models (LLMs) with a concise set of expert-crafted strategies, such as semantic expansion and entity disambiguation, substantially improves retrieval effectiveness on challenging benchmarks, including HotpotQA, FEVER, NFCorpus, and SciFact. Building on this insight, we introduce the Strategy-Adaptive Generation Engine (SAGE), which operationalizes these strategies in an RL framework. SAGE introduces two novel reward shaping mechanisms-Strategic Credit Shaping (SCS) and Contrastive Reward Shaping (CRS)-to deliver more informative learning signals. This strategy-guided approach not only achieves new state-of-the-art NDCG@10 results, but also uncovers a compelling emergent behavior: the agent learns to select optimal strategies, reduces unnecessary exploration, and generates concise rewrites, lowering inference cost without sacrificing performance. Our findings demonstrate that strategy-guided RL, enhanced with nuanced reward shaping, offers a scalable, efficient, and more interpretable paradigm for developing the next generation of robust information retrieval systems.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.