Align Documents to Questions: Question-Oriented Document Rewriting for Retrieval-Augmented Generation

TL;DR

QREAM is a style-controlled document rewriter that aligns retrieved evidence with question-oriented style, improving factual grounding and relevance in retrieval-augmented generation systems.

Contribution

It introduces a novel two-stage framework for rewriting retrieved documents to better serve LLMs, with a lightweight distilled model for efficient deployment.

Findings

QREAM improves RAG performance by up to 8% relative.

The framework enhances factual consistency and question relevance.

QREAM integrates seamlessly as a plug-and-play module.

Abstract

Retrieval-Augmented Generation (RAG) enhances the factuality of Large Language Models (LLMs) by incorporating retrieved documents and/or generated context. However, LLMs often exhibit a stylistic bias when presented with mixed contexts, favoring fluent but hallucinated generated content over factually grounded yet disorganized retrieved evidence. This phenomenon reveals that the utility of retrieved information is bottlenecked by its presentation. To bridge this gap, we propose QREAM, a style-controlled rewriter that aligns retrieved documents with a question-oriented style while preserving facts, better for LLM readers to utilize. Our framework consists of two stages: (1) QREAM-ICL, which uses stylistic seeds to guide iterative rewriting exploration; and (2) QREAM-FT, a lightweight student model distilled from denoised ICL outputs. QREAM-FT employs dual-criteria rejection sampling, filtering based on answer correctness and factual consistency to ensure high-quality supervision. QREAM seamlessly integrates into existing RAG pipelines as a plug-and-play module. Experiments demonstrate that QREAM consistently enhances advanced RAG pipelines, yielding up to 8% relative improvement with negligible latency overhead, effectively balancing question relevance with factual grounding.