A Multichannel Biomedical Named Entity Recognition Model Based on Multitask Learning and Contextualized Word Representations

Table 4

Performance comparison of each representations.

JNLPBA	Precision	Recall	F1-score		NCBI-disease	Precision	Recall	F1-score
Ours	72.6	79.6	76.0	—	Ours	88.2	89.2	88.7	—
ELMo	69.8	76.8	73.1	2.9	ELMo	83.8	85.4	84.6	4.1
Char	66.9	71.9	69.3	6.7	Char	83.7	80.6	82.1	6.6
Word	69.9	75.5	72.6	3.4	Word	84.2	80.9	82.5	6.2
ELMo+Char	71.5	76.4	73.8	2.2	ELMo+Char	86.0	85.5	85.7	3.0
Word+Char	68.9	77.6	73.0	3.0	Word+Char	84.2	85.1	84.7	4.0
Word+ELMo	70.1	77.5	73.6	2.4	Word+ELMo	84.5	86.0	85.3	3.4
Word+ELMo+Char	71.4	77.8	74.4	1.6	Word+ELMo+Char	87.2	85.9	86.6	2.1