Research map / Large language models / Retrieval and knowledge
Micro-Macro Retrieval: Reducing Long-Form Hallucination in Large Language Models
Yujie Feng, Jian Li, Zhihan Zhou, Pengfei Xu, Yujia Zhang, xiaoyu li, Xiaohui Zhou, Alan Zhao, Xi Chen, Xiao-Ming Wu
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 micro retrieval over key-information repository to ReSearch-style macro retrieval | ReSearch (Qwen2.5-7B) · Multi-hop QA (4 benchmarks) | EM: HotpotQA 43.52 -> 44.11; MuSiQue 22.30 -> 24.12 | Improved |
| add micro retrieval to ReSearch | ReSearch (Qwen2.5-3B) · HotpotQA | EM: 38.78 -> 38.70 | No clear effect |
| add micro retrieval to ReSearch | ReSearch (Qwen2.5-3B) · MuSiQue | EM: 19.40 -> 20.87 | Improved |
| on-demand micro retrieval instead of one-shot grounding (ablation) | M2R (7B) · MuSiQue | EM: 23.38 -> 24.12 | Improved |
| two-stage curriculum instead of direct joint optimization (ablation) | M2R (3B) · HotpotQA | EM: 13.2 -> 38.7 | Improved |
| two-stage curriculum instead of direct joint optimization (ablation) | M2R (3B) · MuSiQue | EM: 3.9 -> 20.9 | Improved |
| add micro retrieval vs ReSearch on 5 concatenated questions | M2R (5Q) · MuSiQue-5Q | EM: 5.7 -> 8.4 | Improved |
| add micro retrieval vs ReSearch on other model families | M2R · Bamboogle (Llama-3.1-8B) | EM: 43.1 -> 42.9 | No clear effect |
| add micro retrieval vs ReSearch on Mistral-7B | M2R · MuSiQue (Mistral-7B) | EM: 23.7 -> 25.5 | Improved |
| add micro retrieval | M2R · Inference latency (HotpotQA, 3B) | inference time (s): ~4.3 -> ~4.7 | Hurt |
| vary retrieve top-k 3/5/8 (ablation) | M2R · 2Wiki (3B) | EM: 38.3 / 40.1 / 39.4 | No clear effect |
Closest papers by meaning
- Retrieval Head Mechanistically Explains Long-Context Factuality (ICLR 2025 · similarity 0.88)
- Teaching Language Models to Hallucinate Less with Synthetic Tasks (ICLR 2024 · similarity 0.88)
- Soft Reasoning: Navigating Solution Spaces in Large Language Models through Controlled Embedding Exploration (ICML 2025 · similarity 0.88)
- Coarse-to-Fine Highlighting: Reducing Knowledge Hallucination in Large Language Models (ICML 2024 · similarity 0.88)
- Improving Factuality and Reasoning in Language Models through Multiagent Debate (ICML 2024 · similarity 0.87)
- T1: Advancing Language Model Reasoning through Reinforcement Learning and Inference Scaling (ICML 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.
