Table 3 Comparison with previous deep learning based methods
From: Improving biomedical named entity recognition with syntactic information
Methods | BC2GM | JNLPBA | BC5CDR-chemical | NCBI-disease | LINNAEUS | Species-800 |
|---|---|---|---|---|---|---|
biLSTM + pre-trained embeddings [12] | 78.57 | 77.25 | 91.05 | 84.64 | \(\mathit{94} .\mathit{13}\) | 73.11 |
biLSTM + attentions [23] | – | – | 92.57 | – | – | – |
biLSTM + multi-task learning [43] | 80.74 | 73.52 | - | 86.14 | – | – |
biLSTM + pre-training [31] | 81.69 | 75.03 | – | 87.34 | – | – |
biLSTM + transfer learning [10] | 78.66 | – | 91.64 | 84.72 | 93.54 | 74.98 |
biLSTM + model ensemble [48] | 79.73 | \(\mathit{78} .\mathit{58}\) | 93.31 | 86.36 | – | – |
SciBERT [3] | – | 77.28 | – | 88.57 | – | – |
BERT [19] | 81.79 | 74.94 | 91.16 | 85.63 | 87.60 | 71.63 |
BioBERT (Base) [19] | 84.72 | 77.49 | 93.47 | 89.71 | 88.24 | 75.31 |
BioBERT (Large) [19] | 85.01 | – | – | 88.79 | – | – |
BioBERT (Base) + DR (\({\mathcal {M}}\)) | 84.92 | 77.72 | 94.00 | \(\mathit{90} .\mathit{08}\) | 88.79 | 76.21 |
BioBERT (Large) + DR (\({\mathcal {M}}\)) | \(\mathit{85} .\mathit{29}\) | 77.83 | \(\mathit{94} .\mathit{22}\) | 89.63 | 89.24 | \(\mathit{76} .\mathit{33}\) |