Table 1.

Comparison of the study results at baseline.

Model	Positive			Negative			Overall			AUC-ROCa
	Precision	Recall	F-score	Precision	Recall	F-score	Precision	Recall	F-score
SVMb+BOWc	0.848	0.883	0.865	0.878	0.841	0.859	0.862	0.862	0.862	0.921
SVM+BOW+UMLSd concept	0.891	0.887	0.889	0.870	0.886	0.878	0.886	0.886	0.886	0.934
Autoencoder	0.861	0.855	0.858	0.856	0.862	0.859	0.859	0.859	0.859	0.920
Autoencoder+pretrained word embedding	0.875	0.869	0.872	0.870	0.876	0.873	0.872	0.872	0.872	0.926
CNNe	0.908	0.877	0.892	0.879	0.910	0.894	0.893	0.893	0.893	0.938
CNN+pretrained word embedding	0.930	0.911	0.920	0.912	0.931	0.921	0.921	0.921	0.921	0.946
HCLAf	0.954	0.912	0.932	0.925	0.961	0.943	0.938	0.938	0.938	0.957
CNN for negation bleeding	N/Ag	N/A	N/A	0.820	0.820	0.820	N/A	N/A	N/A	N/A
HCLA for negation bleeding	N/A	N/A	N/A	0.860	0.860	0.860	N/A	N/A	N/A	N/A

^aAUC-ROC: area under the receiver operating characteristic curve.

^bSVM: support vector machines.

^cBOW: bag of words.

^dUMLS: unified medical language system.

^eCNN: convolutional neural network.

^fHCLA: hybrid convolutional neural network and long short-term memory autoencoder.

^gN/A: not applicable.