Research map / Large language models / Retrieval and knowledge
Counterfactual Reasoning for Retrieval-Augmented Generation
Huaiyu Qin, Chunyu Wei, Yueguo Chen, Yunhai 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 causal verification (causal discrimination score) (ablation) | CF-RAG · HotpotQA | exact match (%): 73.19 -> 88.58 | Improved |
| add causal verification (ablation) | CF-RAG · PopQA | exact match (%): 63.47 -> 73.57 | Improved |
| add counterfactual exploration (ablation) | CF-RAG · HotpotQA | exact match (%): 78.52 -> 88.58 | Improved |
| add counterfactual exploration (ablation) | CF-RAG · PopQA | exact match (%): 69.03 -> 73.57 | Improved |
| add evidence division (stratified parallel drafts) (ablation) | CF-RAG · HotpotQA | exact match (%): 84.29 -> 88.58 | Improved |
| add evidence division (ablation) | CF-RAG · PopQA | exact match (%): 67.29 -> 73.57 | Improved |
| use balanced causal weight lambda=0.4 instead of 0 or 1 (ablation) | CF-RAG · HotpotQA | exact match (%): 84.00 / 80.00 -> 88.58 | Improved |
| increase number of parallel drafts to M=3 from M=1 (ablation) | CF-RAG · HotpotQA | exact match (%): 82.23 -> 88.58 | Improved |
| increase counterfactual queries beyond N=3 (ablation) | CF-RAG · HotpotQA | exact match (%): 88.58 -> 86.29 (N=5) | Hurt |
| use CF-RAG instead of Speculative-RAG under 16 adversarial distractors (ablation) | Speculative-RAG · HotpotQA | exact match (%): 20.64 -> 60.57 | Improved |
| use CF-RAG (Llama-3-8B) instead of best baseline (ablation) | Speculative-RAG · HotpotQA | exact match (%): 49.00 -> 88.58 | Improved |
| use CF-RAG instead of Self-CRAG / Speculative-RAG | best baseline · PubHealth | exact match (%): 76.60 -> 83.36 | Improved |
Closest papers by meaning
- Retrieval is Not Enough: Enhancing RAG through Test-Time Critique and Optimization (NeurIPS 2025 · similarity 0.89)
- Worse than Zero-shot? A Fact-Checking Dataset for Evaluating the Robustness of RAG Against Misleading Retrievals (NeurIPS 2025 · similarity 0.88)
- StructRAG: Boosting Knowledge Intensive Reasoning of LLMs via Inference-time Hybrid Information Structurization (ICLR 2025 · similarity 0.87)
- RAGChecker: A Fine-grained Framework for Diagnosing Retrieval-Augmented Generation (NeurIPS 2024 · similarity 0.87)
- Improving Consistency in Retrieval Augmented Systems With Group Similarity Rewards (NeurIPS 2026 · similarity 0.87)
- Speculative RAG: Enhancing Retrieval Augmented Generation through Drafting (ICLR 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.
