TCMNER and PubMed: A Novel Chinese Character-Level-Based Model and a Dataset for TCM Named Entity Recognition

Table 5

The performances of different models in the test set of publications and the entire medical records.

%	Publications (test)			Medical records
	P	R	F1	P	R	F1
BiLSTM-CRF	60.8	30.6	40.7	61.4	48.3	54.1
BERT-CRF	58.7	54.2	56.4	54.7	60.5	57.4
BERT-BiLSTM	93	91.6	90.3	85.5	88.1	86.8
BERT-BiLSTM-CRF	75.4	73.1	74.2	69	75.2	71.9
RoBERTa-BiLSTM	88.8	92.6	90.7	86.3	90.1	88.2
RoBERTa-c	92.6	96.7	94.6	90.4	92.3	91.3