Table 5.

Results of using different NLP toolkits on the BC5CDR-chemical dataset

	BioBERT-base		BioBERT-large
	F1	$\sigma$	F1	$\sigma$
Baseline	93.50	0.10	93.90	0.31
Stanford CoreNLP Toolkits
PL ($\mathcal{M}$)	93.73	0.19	94.05	0.23
DR ($\mathcal{M}$)	$\mathit{93.78}$	0.18	94.05	0.10
spaCy
PL ($\mathcal{M}$)	93.69	0.12	$\mathit{94.06}$	0.10
DR ($\mathcal{M}$)	93.71	0.12	93.97	0.13

The experimental results [the average F1 scores and the standard deviation ($\sigma$)] of our method with KVMN ($\mathcal{M}$) using different NLP toolkits (i.e., Stanford CoreNLP Toolkits and spaCy) to obtain POS labels (PL) and dependency relations (DR). The results of baseline methods without using any syntactic information are also reported for reference