From: Few-shot relation classification by context attention-based prototypical networks with BERT
Model | 5 way 5 shot | 5 way 10 shot | 10 way 5 shot | 10 way 10 shot |
---|---|---|---|---|
Prototypical networks(CNN) | 85.57 ± 0.14 | 88.17 ± 0.10 | 75.01 ± 0.16 | 78.50 ±0.11 |
Proto_CATT(CNN) | 87.48 ± 0.12 | 89.28 ± 0.08 | 77.46 ±0.13 | 80.39 ± 0.14 |
Proto_BERT(fine-tuning) | 91.86 ± 0.02 | 93.34 ± 0.04 | 85.44 ± 0.03 | 87.51 ± 0.04 |
Proto_BERT(CNN) | 93.54 ± 0.05 | 94.68 ± 0.04 | 88.85 ± 0.06 | 90.18 ± 0.05 |
Proto_CATT_BERT(CNN) | 94.86 ± 0.04 | 95.74 ± 0.05 | 90.01 ± 0.04 | 91.60 ± 0.03 |