Improving deep learning method for biomedical named entity recognition by using entity definition information

BMC Bioinformatics

Table 3 Results on PharmaCoNER Corpus

Models	Features	Precision	Recall	F1-score
Xiong et al. [44]	Yes	0.9123	0.9088	0.9105
Stoeckel et al. [47]	No	0.9079	0.9030	0.9052
Sun et al. (2019) [48]	No	0.9046	0.8806	0.8924
Lange et al. [49]	Yes	0.8895	0.8827	0.8861
Hakala et al. [50]	No	0.8758	0.8719	0.8738
Lahuerta et al. [51]	No	0.9022	0.8366	0.8682
Sohrab et al. [52]	Yes	0.8688	0.8665	0.8676
MRC_rule		0.915	0.9055	0.9109
MRC_guideline		0.9225	0.9050	0.9137*
SOne (w/o entity definition)		0.9158	0.8974	0.9065
SOne_rule		0.9153	0.9088	0.912
SOne_guideline		0.9135	0.9121	0.9128
SOne_w2v		0.9167	0.9023	0.9094

The method with the highest F-score among all methods is highlighted in bold
* Compared with the model without any feature, this is a significant improvement (t-test < 0.05)

ISSN: 1471-2105