Biomedical named entity recognition using deep neural networks with contextual information

BMC Bioinformatics

Table 2 Comparison of performance for comparative methods on the NCBI, GM, and CDR corpora using strict and partial matching and IOB tag matching

Model		p	r	f	p	r	f	p	r	f
Strict matching		NCBI			GM			CDR
BiLSTM		78.91	82.60	80.71	72.22	72.44	72.33	83.56	80.26	81.88
BiLSTM-CRF		82.19	84.58	83.37	80.79	79.81	80.30	87.52	83.58	85.50
GRAM-CNN		84.45	83.92	84.18	80.23	78.83	79.53	86.08	85.49	85.79
BERT		81.07	80.73	80.90	81.72	81.59	81.65	86.21	85.23	85.72
CLSTM	word level	85.94	84.69	85.31	81.00	80.77	80.89	87.23	85.51	86.36
	character level	85.40	84.06	84.72	81.09	80.38	80.73	87.19	84.69	85.92
	word+char levels	84.73	86.67	85.68	81.75	81.14	81.44	87.25	85.66	86.44
Partial matching		NCBI			GM			CDR
BiLSTM		86.67	90.73	88.65	87.98	88.25	88.11	91.14	87.54	89.30
BiLSTM-CRF		91.19	93.85	92.51	93.18	92.04	92.61	94.27	90.00	92.08
GRAM-CNN		94.36	93.78	94.07	93.09	91.47	92.27	92.47	91.83	92.15
BERT		88.39	88.02	88.20	92.65	92.51	92.58	91.82	90.77	91.29
CLSTM	word level	93.66	92.29	92.97	92.81	92.54	92.67	93.60	91.74	92.66
	character level	93.76	92.29	93.02	93.05	92.25	92.65	93.42	91.59	92.49
	word+char levels	93.71	93.13	93.42	93.35	92.65	93.00	93.48	91.77	92.62
IOB tag matching		NCBI			GM			CDR
BiLSTM		84.56	88.03	86.26	84.23	81.48	82.83	89.81	78.68	83.87
BiLSTM-CRF		84.13	88.32	86.18	88.34	84.47	86.36	90.54	81.34	85.69
GRAM-CNN		88.73	86.59	87.65	87.75	84.09	85.89	89.72	83.03	86.24
BERT		88.42	83.15	85.70	89.50	86.26	87.85	88.69	85.01	86.81
CLSTM	word level	89.18	89.01	89.10	88.99	84.89	86.89	89.99	83.19	86.45
	character level	88.21	88.47	88.34	87.72	85.21	86.45	89.95	83.19	86.43
	word+char levels	89.98	87.74	88.84	87.13	86.88	87.00	90.56	83.41	86.83

ISSN: 1471-2105