Research map / Large language models / Retrieval and knowledge
REAP: Enhancing RAG with Recursive Evaluation and Adaptive Planning for Multi-Hop Question Answering
Yijie Zhu, Haojie Zhou, Wanting Hong, Tailin Liu, Ning Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Re-Planner sub-module (ablation) | REAP · HotpotQA, MuSiQue, Bamboogle | F1: 64.9 -> 68.0 (HotpotQA) | Improved |
| add reasoning and verification mechanism in Fact Extractor (ablation) | REAP · Bamboogle | F1: 60.8 -> 65.2 | Improved |
| add clue feedback mechanism (ablation) | REAP · HotpotQA, 2Wiki | F1: 64.6 -> 68.0 (HotpotQA); 76.5 -> 79.6 (2Wiki) | Improved |
| multi-task fine-tuning of planning modules instead of separate models (ablation) | REAP · 2Wiki, Bamboogle | F1: 77.2 -> 79.6 (2Wiki) | Improved |
| multi-task design versus naive joint fine-tuning (ft-all) (ablation) | REAP · 2Wiki, MuSiQue | F1: 78.4 -> 79.6 (2Wiki) | Improved |
| replace Plan Updater model with fine-tuned Llama-3.2-1B (ablation) | REAP · HotpotQA, 2Wiki, MuSiQue, Bamboogle | F1: 68.0 -> 69.0 (HotpotQA); 65.2 -> 61.8 (Bamboogle) | Mixed |
| iterative plan and fact extraction vs fine-tuned single-step RAG | REAP vs FT-Standard RAG · HotpotQA | F1: 61.2 -> 68.0 | Improved |
| swap generator from Llama-3.1-8B to Llama-3.1-70B without fine-tuning | REAP · HotpotQA, 2Wiki, MuSiQue, Bamboogle | F1: REAP70B 65.5, 70.0, 37.2, 61.6 vs best baselines lower | Improved |
Closest papers by meaning
- LiR3AG: A Lightweight Rerank Reasoning Strategy Framework for Retrieval-Augmented Generation (AAAI 2026 · similarity 0.92)
- FrugalRAG: Less is More in RL Finetuning for Multi-hop Question Answering (ICLR 2026 · similarity 0.90)
- OPERA: A Reinforcement Learning--Enhanced Orchestrated Planner-Executor Architecture for Reasoning-Oriented Multi-Hop Retrieval (AAAI 2026 · similarity 0.90)
- Cooperative Retrieval-Augmented Generation for Question Answering: Mutual Information Exchange and Ranking by Contrasting Layers (NeurIPS 2025 · similarity 0.90)
- Tree-of-Reasoning Question Decomposition for Complex Question Answering with Large Language Models (AAAI 2024 · similarity 0.89)
- Adapting to Non-Stationary Environments: Multi-Armed Bandit Enhanced Retrieval-Augmented Generation on Knowledge Graphs (AAAI 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
