Table 6.

Performance comparison of pre-trained language models

Model	Accuracy	Precision	Recall	F1-score
BERT [20]a	*0.849 **(0.003)	0.817 (0.010)	0.822 (0.019)	0.818 (0.011)
BioBERT [31]b	0.861 (0.008)	0.835 (0.017)	0.846 (0.015)	0.839 (0.011)
PubMedBERT[34]c	0.865 (0.014)	0.833 (0.020)	0.849 (0.009)	0.839 (0.015)
RoBERTa [35]d	0.862 (0.009)	0.835 (0.018)	0.837 (0.009)	0.835 (0.010)
SciBERT [32]e	0.862 (0.010)	0.836 (0.017)	0.843 (0.013)	0.838 (0.013)

The best scores are in bold

^*Mean

^**Standard deviation

^aBert-base-uncased, Accessed July 20, 2022, Available from: https://huggingface.co/bert-base-uncased

^bBiobert-base-cased-v1.2, Accessed July 20, 2022, Available from: https://huggingface.co/dmis-lab/biobert-base-cased-v1.2

^cBiomedNLP-PubMedBERT-base-uncased-abstract-fulltext, Accessed July 20, 2022, Available from: https://huggingface.co/microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract-fulltext

^dRoberta-base, Accessed July 20, 2022, Available from: https://huggingface.co/roberta-base

^eScibert_scivocab_uncased, Accessed July 20, 2022, Available from: https://huggingface.co/allenai/scibert_scivocab_uncased