Table 4:

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

Model	Strict Criterion			Relaxed Criterion
	Precision	Recall	F1	Precision	Recall	F1
BERT	0.7618	0.7919	0.7765	0.8972	0.9208	0.9089
BERT-MIMIC	0.7517	0.7949	0.7727	0.8895	0.9290	0.9088
BLUEBERT	0.7550	0.7948	0.7744	0.8933	0.9280	0.9103
ALBERT	0.7248	0.7671	0.7454	0.8644	0.9007	0.8822
ALBERT-MIMIC	0.7422	0.7682	0.7550	0.8820	0.8974	0.8896
RoBERTa	0.7610	0.8061	0.7829	0.8908	0.9333	0.9115
RoBERTa-MIMIC	0.7601	0.8026	0.7808	0.8918	0.9331	0.9120
RoBERTa-MIMIC-Trial	0.7680	0.8027	0.7849	0.9028	0.9290	0.9157
ELECTRA	0.7444	0.7910	0.7670	0.8840	0.9254	0.9042
ELECTRA-MIMIC	0.7683	0.7903	0.7792	0.9034	0.9185	0.9109
Att-BiLSTM-CRF	0.6799	0.7199	0.6993	0.8692	0.9039	0.8862