Research map / Large language models / Retrieval and knowledge
Promptagator: Few-shot Dense Retrieval From 8 Examples
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 8-shot prompt-based query generation with FLAN instead of zero-shot prompt | Promptagator retriever · BEIR 11 datasets | nDCG@10 avg: +2.3 (45.5 to 47.8) | Helped |
| Add round-trip consistency filtering (K=1) using generated data only (ablation) | Promptagator few-shot retriever · BEIR 11 datasets | nDCG@10 avg: about +2.5 average; improves 8 of 11 datasets | Helped |
| Add consistency filtering using generated data alone (ablation) | Promptagator few-shot retriever · NFCorpus and SciFact | nDCG@10 | Hurt |
| Add a reranker trained on the same generated data (top 200 reranking) | Promptagator dual encoder · BEIR 11 datasets | nDCG@10 avg: about +5 points (47.8 to 52.8) | Helped |
| Use FEVER few-shot prompt instead of Climate-FEVER examples (ablation) | Promptagator few-shot retriever on Climate-FEVER · Climate-FEVER | nDCG@10: 16.8 to 24.0 | Helped |
| Train FLAN without NQ and Quora data (ablation) | Promptagator few-shot retriever · BEIR 11 datasets | nDCG@10 avg: -1.5 (48.5 to 47.0) | Hurt |
| Use FLAN-generated queries rather than T5 NQ-QGen queries (ablation) | dual encoder trained on generated queries · BEIR (zero-shot prompt comparison) | nDCG@10 | Helped |
Closest papers by meaning
- Promptriever: Instruction-Trained Retrievers Can Be Prompted Like Language Models (ICLR 2025 · similarity 0.90)
- PQPP: A Joint Benchmark for Text-to-Image Prompt and Query Performance Prediction (CVPR 2025 · similarity 0.88)
- Learning to Rank in Generative Retrieval (AAAI 2024 · similarity 0.88)
- Attention in Large Language Models Yields Efficient Zero-Shot Re-Rankers (ICLR 2025 · similarity 0.87)
- The Devil is in the Prompts: Retrieval-Augmented Prompt Optimization for Text-to-Video Generation (CVPR 2025 · similarity 0.87)
- Generative Retrieval Meets Multi-Graded Relevance (NeurIPS 2024 · 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.