Table 4.
A comparison of LSTM-CRFs and BERT for subtask 1 (concept extraction).
| Model and concept | Precision | Recall | F1 score | |
| LSTM-CRFsa,b |
|
|
|
|
|
|
Family member | 0.8480 | 0.7686 | 0.8069 |
| Observation | 0.7382 | 0.8342 | 0.7833 | |
| LSTM-CRFs-EN |
|
|
|
|
|
|
Family member | 0.8451 | 0.7868 | 0.8149 |
| Observation | 0.7685 | 0.7953 | 0.7817 | |
| BERTc-ner |
|
|
|
|
|
|
Family member | 0.8059 | 0.8072 | 0.8066 |
| Observation | 0.8061 | 0.8127 | 0.8094 | |
| BERT-ner-EN |
|
|
|
|
|
|
Family member | 0.8294 | 0.8229 | 0.8261 |
| Observation | 0.8306 | 0.8178 | 0.8241 | |
aLSTM: long short-term memory.
bCRFs: conditional random fields.
cBERT: bidirectional encoder representations from transformers.