Table 3 Overall performance of the PICO entity recognition models on the test set
From: PICO entity extraction for preclinical animal literature
Model | Weight | F1 | Recall | Precision |
|---|---|---|---|---|
BiLSTM | – | 43.5 | 38.1 | 50.6 |
BiLSTM-CRF | – | 57.9 | 54.7 | 61.6 |
BERT | Base | 61.3 | 66.3 | 57.1 |
BioBERT | 65.4 | 69.8 | 61.5 | |
PubMed-abs | 70.1 | 73.2 | 67.3 | |
PubMed-full | 69.9 | 73.4 | 66.7 | |
BERT-CRF | Base | 62.1 | 67.2 | 57.8 |
BioBERT | 66.5 | 70.1 | 63.3 | |
PubMed-abs | 68.0 | 71.5 | 64.9 | |
PubMed-full | 67.5 | 70.9 | 64.5 | |
BERT  - BiLSTM  - CRF | Base | 64.6 | 69.5 | 60.3 |
BioBERT | 68.3 | 71.2 | 65.6 | |
PubMed-abs | 67.2 | 70.8 | 64.0 | |
PubMed-full | 68.5 | 72.6 | 64.8 |