Question Answering Retrieval on STaRK synthetically generated
62Hit@1AF-Retriever
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| AF-RetrieverLLM Backbone=GPT OSS 120B, Embedding=text-embedding-3-small, Protocol=Zero-shot, lmax=100, k=202025.05 | 62 | 76.7 | 68.4 | |
| 4StepFocusData Version=10%-version2025.05 | 46.9 | 63.3 | 54.6 | |
| KARProtocol=Zero-shot / One-shot2025.05 | 45 | 62.5 | 53 | |
| AvaTaRProtocol=Zero-shot / One-shot2025.05 | 37.6 | 55.2 | 45.5 | |
| VSS + RerankerData Version=10%-version2025.05 | 34.7 | 55.5 | 43.7 | |
| ReflexionProtocol=Zero-shot / One-shot2025.05 | 32.6 | 51.5 | 41.6 | |
| ReActProtocol=Zero-shot / One-shot2025.05 | 29.5 | 48.7 | 38.1 | |
| BM25Protocol=Zero-shot / One-shot2025.05 | 27.8 | 46.9 | 36.7 | |
| VSS (Ada-002)Embedding=Ada-0022025.05 | 27 | 47.9 | 36.8 | |
| VSS (Multi-Ada)Embedding=Multi-Ada2025.05 | 27 | 49.7 | 37.3 | |
| QA-GNNProtocol=Zero-shot / One-shot2025.05 | 16.1 | 36.8 | 27.2 | |
| DPRProtocol=Zero-shot / One-shot2025.05 | 10.1 | 35 | 21.3 |