Table 6 Overall performance of PICO entity recognition models on the validation set
From: PICO entity extraction for preclinical animal literature
Model | Weight | F1 | Recall | Precision |
|---|---|---|---|---|
BiLSTM | – | 41.7 | 44.2 | 39.5 |
BiLSTM-CRF | – | 58.8 | 56.9 | 61.0 |
BERT | Base | 56.0 | 62.7 | 50.6 |
BioBERT | 64.2 | 69.8 | 59.4 | |
PubMed-abs | 65.0 | 70.5 | 60.2 | |
PubMed-full | 68.1 | 73.0 | 63.8 | |
BERT-CRF | Base | 57.7 | 62.9 | 53.3 |
BioBERT | 65.1 | 70.0 | 60.9 | |
PubMed-abs | 65.5 | 70.9 | 60.9 | |
PubMed-full | 68.0 | 72.8 | 63.7 | |
BERT  - BiLSTM  - CRF | Base | 60.8 | 66.4 | 56.1 |
BioBERT | 66.0 | 70.0 | 62.5 | |
PubMed-abs | 68.1 | 73.3 | 63.5 | |
PubMed-full | 68.0 | 72.8 | 63.8 |