Skip to main content
. Author manuscript; available in PMC: 2021 Aug 18.
Published in final edited form as: ACM BCB. 2021 Aug;2021:49. doi: 10.1145/3459930.3469560

Table 3:

Performance of the transformer-based models vs. the baseline Att-BiLSTM-CRF model on Chia.

Model Strict Criterion Relaxed Criterion
Precision Recall F1 Precision Recall F1
BERT 0.6052 0.6653 0.6339 0.7646 0.8132 0.7882
BERT-MIMIC 0.5934 0.6749 0.6316 0.7559 0.8228 0.7879
BLUEBERT 0.6244 0.6634 0.6433 0.7819 0.8033 0.7925
ALBERT 0.6007 0.6488 0.6238 0.7715 0.8020 0.7864
ALBERT-MIMIC 0.6329 0.6475 0.6401 0.7871 0.7818 0.7845
RoBERTa 0.6312 0.6818 0.6556 0.7715 0.8155 0.7929
RoBERTa-MIMIC 0.6158 0.6766 0.6448 0.7711 0.8175 0.7936
RoBERTa-MIMIC-Trial 0.6209 0.6993 0.6578 0.7662 0.8333 0.7984
ELECTRA 0.5749 0.6498 0.6101 0.7369 0.8013 0.7678
ELECTRA-MIMIC 0.6086 0.6723 0.6389 0.7661 0.8149 0.7897
Att-BiLSTM-CRF 0.3586 0.3896 0.3735 0.7064 0.7344 0.7201