GRAPE: Let GRPO Supervise Query Rewriting by Ranking for Retrieval
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2025
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866917470260953088 |
|---|---|
| author | Zhang, Zhaohua Zhuo, Jianhuan Chen, Muxi Zhao, Chenchen Jiang, Wenyu Jiang, Tianwen Chen, Mingyang Yutang Xiao, Qiuyong Zhang, Jihong Su, Zhixun |
| author_facet | Zhang, Zhaohua Zhuo, Jianhuan Chen, Muxi Zhao, Chenchen Jiang, Wenyu Jiang, Tianwen Chen, Mingyang Yutang Xiao, Qiuyong Zhang, Jihong Su, Zhixun |
| contents | The CLIP model has established itself as a cornerstone of large-scale retrieval systems. However, its performance often degrades under distributional shifts such as multilingual, long-form, or multimodal queries. To avoid the prohibitive costs associated with retriever retraining or corpus re-embedding, we propose GRAPE (Grouped Ranking-Aware Policy Optimization Enhancement), a plug-and-play approach that leverages LLM-based query rewriting to bridge these gaps. Unlike existing methods that lack explicit supervision, GRAPE integrates ranking signals into the rewriting LLM via Grouped Relative Policy Optimization (GRPO), ensuring rewritten queries are better aligned with the frozen retriever's latent distribution. Crucially, we identify a score inflation phenomenon in naive similarity-based finetuning - where irrelevant candidates receive indiscriminately high scores - and mitigate it with a novel corpus-relative ranking-based reward. Extensive experiments across multilingual (Flickr30k-CN, CVLUE, XM3600), long-form (Wikipedia), and multimodal (CIRR) benchmarks demonstrate that GRAPE consistently improves performance, achieving an average gain of 4.9% in Recall@10 without any modification to the underlying retriever. The code is available at https://github.com/mogulzhang/GRAPE. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_23370 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | GRAPE: Let GRPO Supervise Query Rewriting by Ranking for Retrieval Zhang, Zhaohua Zhuo, Jianhuan Chen, Muxi Zhao, Chenchen Jiang, Wenyu Jiang, Tianwen Chen, Mingyang Yutang Xiao, Qiuyong Zhang, Jihong Su, Zhixun Computer Vision and Pattern Recognition The CLIP model has established itself as a cornerstone of large-scale retrieval systems. However, its performance often degrades under distributional shifts such as multilingual, long-form, or multimodal queries. To avoid the prohibitive costs associated with retriever retraining or corpus re-embedding, we propose GRAPE (Grouped Ranking-Aware Policy Optimization Enhancement), a plug-and-play approach that leverages LLM-based query rewriting to bridge these gaps. Unlike existing methods that lack explicit supervision, GRAPE integrates ranking signals into the rewriting LLM via Grouped Relative Policy Optimization (GRPO), ensuring rewritten queries are better aligned with the frozen retriever's latent distribution. Crucially, we identify a score inflation phenomenon in naive similarity-based finetuning - where irrelevant candidates receive indiscriminately high scores - and mitigate it with a novel corpus-relative ranking-based reward. Extensive experiments across multilingual (Flickr30k-CN, CVLUE, XM3600), long-form (Wikipedia), and multimodal (CIRR) benchmarks demonstrate that GRAPE consistently improves performance, achieving an average gain of 4.9% in Recall@10 without any modification to the underlying retriever. The code is available at https://github.com/mogulzhang/GRAPE. |
| title | GRAPE: Let GRPO Supervise Query Rewriting by Ranking for Retrieval |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2509.23370 |