Question Answering 벤치마크
Question Answering on ConditionalQA
Conditional (answers) 낮을수록 좋음
- 2020-04-17 — ETC-Pipeline: Conditional (answers) 39.4
| Rank | Model | Conditional (answers) | Conditional (w/ conditions) | Overall (answers) | Overall (w/ conditions) | Paper | Code | Year |
|---|---|---|---|---|---|---|---|---|
| 1 | FiD | 45.2 / 49.7 | 4.7 / 5.8 | 44.4 / 50.8 | 35.0 / 40.6 | Leveraging Passage Retrieval with Generative Models for Open Domain Question Answering | jhyuklee/DensePhrases · princeton-nlp/DensePhrases · facebookresearch/FiD · +5 | 2020 |
| 2 | DocHopper | 42.0 / 46.4 | 3.1 / 3.8 | 40.6 / 45.2 | 31.9 / 36.0 | Iterative Hierarchical Attention for Answering Complex Questions over Long Documents | 2021 | |
| 3 | ETC-Pipeline | 39.4 / 41.8 | 2.5 / 3.4 | 35.6 / 39.8 | 26.9 / 30.8 | ETC: Encoding Long and Structured Inputs in Transformers | google-research/google-research · google-research/longt5 | 2020 |