Research map / Large language models / Retrieval and knowledge
CiteGuard: Conformal False-Discovery Control for Faithful Retrieval-Augmented Generation
Xiangyu Jiang
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 |
|---|---|---|---|
| replace Self-RAG with CITEGUARD (BH) | RAG pipeline · FEVER dev | citation precision: 85.2 -> 90.3 | Improved |
| replace Self-RAG with CITEGUARD (BH) | RAG pipeline · NQ human | EM@Acc: 44.6 -> 49.2 | Improved |
| use BY instead of Self-RAG | RAG pipeline · NQ human | FDR: 0.204 -> 0.094 | Improved |
| use BY correction instead of BH | CITEGUARD · FEVER dev | FDR: 0.097 -> 0.073 | Improved |
| compare at matched ~27% abstention against tuned Self-RAG | CITEGUARD (BH) · FEVER | FDR: 0.128 -> 0.097 | Improved |
| use Contriever instead of BM25 retriever (ablation) | CITEGUARD (BH) · FEVER | abstention rate: 34.5 -> 22.1 | Improved |
| use DPR instead of BM25 retriever (ablation) | CITEGUARD (BH) · FEVER | coverage: 86.8 -> 92.3 | Improved |
| use atomic-fact claim extraction instead of full-sentence (ablation) | CITEGUARD (BH) · NQ human | FDR: 0.104 -> 0.098 | Improved |
| use atomic-fact claim extraction instead of full-sentence (ablation) | CITEGUARD (BH) · NQ human | abstention rate: 31.2 -> 36.4 | Hurt |
| add 20% label noise to calibration (ablation) | CITEGUARD (BH) · FEVER | FDR: 0.097 -> 0.128 | Hurt |
| test on shifted 2023 Wikipedia without recalibration | CITEGUARD (BH) · FEVER | FDR: 0.097 -> 0.118 | Hurt |
| change generator from FiD to Llama-3 without recalibration | CITEGUARD (BH) · NQ | FDR: +4.8 pp | Hurt |
Closest papers by meaning
- Worse than Zero-shot? A Fact-Checking Dataset for Evaluating the Robustness of RAG Against Misleading Retrievals (NeurIPS 2025 · similarity 0.90)
- Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection (ICLR 2024 · similarity 0.88)
- Retrieval is Not Enough: Enhancing RAG through Test-Time Critique and Optimization (NeurIPS 2025 · similarity 0.88)
- ParamMute: Suppressing Knowledge-Critical FFNs for Faithful Retrieval-Augmented Generation (NeurIPS 2025 · similarity 0.87)
- Large language model validity via enhanced conformal prediction methods (NeurIPS 2024 · similarity 0.86)
- ClashEval: Quantifying the tug-of-war between an LLM’s internal prior and external evidence (NeurIPS 2024 · similarity 0.86)
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.
