. Author manuscript; available in PMC: 2020 Jul 13.

Published in final edited form as: J Biomed Inform. 2019 Feb 10;91:103123. doi: 10.1016/j.jbi.2019.103123

Table 4:

The results of SVM experiments with various feature combinations. The best performance for individual metrics are underlined. The evaluation is based on 10-fold cross-validation.

Experiment	Overall		Per Category
Experiment	Accu.	MacroF₁	Cat	Pr.	Rec.	F₁
Results from Xu et al. [13]	0.870	0.719	POS	0.823	0.644	0.723
			NEG	0.711	0.399	0.511
			NEU	0.886	0.966	0.924
Reproducing Xu et al. (ngram+sent+struct)	0.870	0.717	POS	0.835	0.618	0.710
			NEG	0.717	0.403	0.516
			NEU	0.883	0.971	0.925
ngraml+sent+struct	0.873	0.720	POS	0.854	0.631	0.726
			NEG	0.709	0.396	0.508
			NEU	0.884	0.973	0.927
ngraml+sent+struct+dep+rule	0.878	0.735	POS	0.846	0.664	0.744
			NEG	0.733	0.419	0.533
			NEU	0.891	0.970	0.929
ngraml+sent+struct+posgram+rule	0.876	0.738	POS	0.808	0.672	0.734
			NEG	0.739	0.442	0.553
			NEU	0.895	0.963	0.927
ngraml_pos+posgram+sent+struct+dep+rule	0.877	0.741	POS	0.828	0.667	0.739
			NEG	0.751	0.442	0.556
			NEU	0.891	0.966	0.927
ngraml_pos+sent+struct+rule	0.878	0.737	POS	0.836	0.660	0.737
			NEG	0.724	0.435	0.544
			NEU	0.893	0.969	0.930