Research map / Large language models / Instruction and fine-tuning
FineRef: Fine-Grained Error Reflection and Correction for Long-Form Generation with Citations
Yixing Peng, Licheng Zhang, Shancheng Fang, Yi Liu, Peijian Gu, Quan 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 reflection step (ablation) | FineRef (Mistral-Orca-7B) · ALCE-ASQA | citation F1: 79.4 -> 85.2 | Improved |
| add irrelevance error type to reflection (ablation) | FineRef (Mistral-Orca-7B) · ALCE-ASQA | EM recall: 39.9 -> 43.4 | Improved |
| use fine-grained per-citation reflection instead of coarse GPT-4 reflection (ablation) | FineRef (Mistral-Orca-7B) · ALCE-ASQA | citation F1: 81.1 -> 85.2 | Improved |
| add RL stage (ablation) | FineRef (Mistral-Orca-7B) · ALCE-ASQA | citation F1: 80.6 -> 85.2 | Improved |
| add correction gain reward (ablation) | FineRef (Mistral-Orca-7B) · ALCE-ASQA | citation F1: 81.6 -> 85.2 | Improved |
| add online self-reflective bootstrapping (ablation) | FineRef (Mistral-Orca-7B) · ALCE-ASQA | citation F1: 81.5 -> 85.2 | Improved |
| apply correction step after first attempt | FineRef (Mistral-Orca-7B) · ALCE-ASQA | citation F1: 77.4 -> 84.7 | Improved |
| FineRef vs CaLF on ASQA | LLaMA2-7B-chat · ALCE-ASQA | citation F1: 70.4 -> 74.9 | Improved |
| FineRef vs CaLF on ELI5 | Mistral-Orca-7B · ALCE-ELI5 | citation F1: 73.1 -> 76.3 | Improved |
| zero-shot domain transfer ELI5->ASQA vs CaLF | Mistral-Orca-7B · ALCE-ASQA | citation F1: 79.5 -> 81.3 | Improved |
| insert distractor passage (noisy retrieval) | CaLF and few-shot FT · ALCE | EM recall | Hurt |
Closest papers by meaning
- Chain-of-Thought Improves Text Generation with Citations in Large Language Models (AAAI 2024 · similarity 0.89)
- SelfCite: Self-Supervised Alignment for Context Attribution in Large Language Models (ICML 2025 · similarity 0.85)
- ClashEval: Quantifying the tug-of-war between an LLM’s internal prior and external evidence (NeurIPS 2024 · similarity 0.85)
- RevisEval: Improving LLM-as-a-Judge via Response-Adapted References (ICLR 2025 · similarity 0.85)
- Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection (ICLR 2024 · similarity 0.85)
- Towards Verifiable Text Generation with Generative Agent (AAAI 2025 · similarity 0.85)
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.
