Table 6.
Comparison of F1 values on different models.
| Model | BM_GBD | Sci ERC | NYT | ACE04 | ACE05 |
|---|---|---|---|---|---|
| Qi Li and Heng Ji (2014) | 23.5* | 12.6* | 45.3* | 45.3 | 49.5 |
| Miwa and Bansal (2016) | 25.4* | 20.0* | 43.4* | 43.2* | 50.1* |
| Katiyar and Cardie (2017) | 15.7* | 18.5 | 48.6* | 45.7 | 53.6 |
| Zheng et al. (2017) | 21.1* | 19.2* | 52 | 48.8* | 57.5 |
| Yu et al. (2021) | 65.1* | 33.9* | 59 | 60.5* | 59.1* |
| Zhong and Chen (2020) | 70.8* | 36.8 | 86.2* | 67 | 62.2 |
| Sci BERT | 71.1* | 33.9* | 88.3* | 60.4* | 58.0* |
| Sci BERT+SelfMix | 74.3* | 34.6* | 90.6* | 64.0* | 61.3* |
| Sci BERT+CRF | 73.4* | 33.8* | 88.3* | 62.7* | 59.2* |
| Sci BERT+L-CRF | 73.5* | 34.5* | 89.8* | 62.4* | 57.0* |
| Sci BERT+SelfMix+L-CRF | 75.5* ∗ | 37.2* | 90.9* | 66.8* | 60.2* |
The results marked with “*” are from our independent experiments, since some researches don’t provide results for related datasets, while those without “*” are directly cited from corresponding papers.
Bold values indicate the best performance for each metric across the models.