Research map / Large language models / Retrieval and knowledge
Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, Douwe Kiela
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 retrieval-augmented generation instead of DPR extractive QA | Natural Questions · Natural Questions | Exact Match: 41.5 -> 44.5 | Improved |
| add retrieval-augmented generation instead of DPR extractive QA | WebQuestions · WebQuestions | Exact Match: 41.1 -> 45.5 | Improved |
| use RAG-Sequence rather than RAG-Token | TriviaQA standard test · TriviaQA (standard test) | Exact Match: 55.2 -> 56.8 | Mixed |
| use RAG-Token rather than RAG-Sequence | Jeopardy question generation · Jeopardy question generation | Q-BLEU-1: 21.4 -> 22.2 | Improved |
| add retrieval-augmented generation | MS-MARCO generation versus BART · MS-MARCO | Rouge-L: 38.2 -> 40.8 | Improved |
| add retrieval-augmented generation | Jeopardy question generation versus BART · Jeopardy question generation | Q-BLEU-1: 19.7 -> 22.2 | Improved |
| add retrieval-augmented generation rather than BART | Jeopardy question generation · Jeopardy question generation | human-rated factuality: 7.1% -> 42.7% | Improved |
| replace learned dense retrieval with fixed BM25 retrieval (ablation) | RAG · Natural Questions (dev) | Exact Match: 44.0 -> 31.8 | Hurt |
| replace learned dense retrieval with fixed BM25 retrieval (ablation) | RAG · FEVER 3-way (dev) | label accuracy: 74.5 -> 75.1 | Improved |
| train by fine-tuning the retriever rather than freezing it (ablation) | RAG · Natural Questions (dev) | Exact Match: 41.2 -> 44.0 | Improved |
| train by fine-tuning the retriever rather than freezing it (ablation) | RAG · WebQuestions (dev) | Exact Match: 37.1 -> 46.5 | Improved |
| add a null-document mechanism (ablation) | RAG · Open-domain tasks | performance | No clear effect |
Closest papers by meaning
- Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection (ICLR 2024 · similarity 0.90)
- Long-Context LLMs Meet RAG: Overcoming Challenges for Long Inputs in RAG (ICLR 2025 · similarity 0.90)
- A Theory for Token-Level Harmonization in Retrieval-Augmented Generation (ICLR 2025 · similarity 0.90)
- From RAG to Memory: Non-Parametric Continual Learning for Large Language Models (ICML 2025 · similarity 0.90)
- Inference Scaling for Long-Context Retrieval Augmented Generation (ICLR 2025 · similarity 0.90)
- One Token Can Help! Learning Scalable and Pluggable Virtual Tokens for Retrieval-Augmented Large Language Models (AAAI 2025 · similarity 0.90)
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.
