Only Ask What You Don't Know: Grounded Delta Planning for Efficient Multi-step RAG

2026-06-21

AI Summary

Existing retrieval-augmented generation (RAG) approaches for multi-hop question answering often suffer from error propagation and redundant reasoning due to iterative retrieval, leading to suboptimal efficiency and accuracy. To address this, this work proposes GDP-RAG, a novel framework that introduces an information-gain-driven planning mechanism. It first anchors known information through an initial retrieval and then generates sub-queries exclusively targeting uncovered knowledge gaps. By integrating evidence-grounded, skeleton-style reasoning trajectories, the method focuses inference on critical missing pieces, enabling efficient multi-step reasoning. Experiments demonstrate that GDP-RAG achieves a state-of-the-art accuracy of 60.63% on HotpotQA, 2WikiMultiHopQA, and MuSiQue while significantly reducing computational cost—requiring only 78% of the resources of PAR-RAG and 32% of those of KnowTrace—thus jointly optimizing both precision and efficiency.

Abstract

Multi-hop question answering remains challenging for Retrieval-Augmented Generation (RAG) because existing approaches either propagate errors across iterative retrieval rounds or over-generate reasoning steps, increasing cost without improving accuracy. We propose Grounded Delta Planning RAG (GDP-RAG), a plan-based framework that targets only the information delta based on three simple design choices: (1) preliminary retrieval to ground planning before execution, (2) a gap-conditioned planning prompt that asks only for missing information, and (3) a skeletal trajectory that pairs each subquery with a Thought capturing evidence from preliminary retrieval and carrying it through to the final answer. GDP-RAG focuses computation on unresolved gaps, yielding concise, reliable reasoning trajectories. Extensive experiments on HotpotQA, 2WikiMultiHopQA, and MuSiQue show that GDP-RAG achieves the highest accuracy (60.63%) among all compared systems while maintaining a cost-of-pass of 0.51, 22% lower than PAR-RAG (0.65) and 68% lower than KnowTrace (1.57), with no method achieving both higher accuracy and lower cost.

Problem

• Multi-hop question answering
• Retrieval-Augmented Generation
• error propagation
• over-generation
• reasoning efficiency

Innovation

• Grounded Delta Planning
• Retrieval-Augmented Generation
• Multi-hop Question Answering
• Gap-conditioned Prompting
• Efficient Reasoning