. 2023 Apr 14;4(4):100729. doi: 10.1016/j.patter.2023.100729

Table 11.

Comparison of PubMedBERT fine-tuned models, scispaCy and Stanza on BLURB named-entity recognition tasks (relaxed entity-level test F1 score—overlap counted as correct)

	scispaCy			Stanza					PubMedBERT
	jnlpba	bc5cdr	max	jnlpba	bc5cdr	ncbi-disease	bc4chemd	max	$B A S E$	$L A R G E$
BC5-chem	7.70	91.42	91.42	–	94.31	–	92.61	94.31	95.18	95.37^∗
BC5-disease	2.09	88.88	88.88	–	92.10	77.79	–	92.10	93.34	93.74^∗
NCBI-disease	12.94	74.64	74.64	–	81.83	93.37	–	93.37	95.22	95.24^∗
BC2GM	68.87	15.92	68.87	–	–	–	–	–	95.56	96.05^∗
JNLPBA	87.25	20.50	87.25	88.33	–	–	–	88.33	88.79	88.81^∗
Mean Score	35.77	58.27	82.21	88.33	89.42	85.58	92.61	92.02	93.62	93.84^∗

^∗

Highest performance for task (row).