Improving the recall of biomedical named entity recognition with label re-correction and knowledge distillation

BMC Bioinformatics

Table 6 Comparison with some state-of-the-art methods

	Methods	CDR chemical F(%)	CDR disease F(%)	CDR both F(%)	NCBI disease F(%)
1	Habibi et al. [5]	91.05	83.49	87.63*	84.44
	Our baseline (BiLSTM-CRF)	91.42	83.59	87.86	83.96
	Our baseline (BioBERT-CRF)	93.69	86.19	90.31	87.47
2	Luo et al. [1]	92.57	–	–	–
2	Dang et al. [18]	93.14	84.68	89.30*	84.41
3	Wang et al. [21]	–	–	88.78	86.14
3	Yoon et al. [22]	92.74	82.61	88.15*	86.36
4	Lee et al. [24]	93.47	87.15	90.60*	89.71
	Our model (BiLSTM-CRF)	94.17	85.69	90.35	85.71
	Our model (BioBERT-CRF)	95.22	87.34	91.64	89.75

The highest scores are highlighted in bold
1: models with word and character features
2: models with additional domain resource features and linguistic features
3: models with multi-task learning
4: models with large-scale unlabeled datasets
*Indicates that the results are calculated by us according to their reported results in chemical and disease

ISSN: 1471-2105