Saved in:
Bibliographic Details
Main Authors: Wang, Teng, Gong, Hailei, Zhang, Changwang, Wang, Jun
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2506.19783
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911078037848064
author Wang, Teng
Gong, Hailei
Zhang, Changwang
Wang, Jun
author_facet Wang, Teng
Gong, Hailei
Zhang, Changwang
Wang, Jun
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2506_19783
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle SAGE: Strategy-Adaptive Generation Engine for Query Rewriting
Wang, Teng
Gong, Hailei
Zhang, Changwang
Wang, Jun
Artificial Intelligence
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.
title SAGE: Strategy-Adaptive Generation Engine for Query Rewriting
topic Artificial Intelligence
url https://arxiv.org/abs/2506.19783