Negation detection in Dutch clinical texts: an evaluation of rule-based and machine learning methods

BMC Bioinformatics

Table 3 Classification results across methods and data sources

Letter category	Prediction method	Precision	Recall	F1
Discharge letters	Rule-based	0.893	0.921	0.906
Discharge letters	BiLSTM	0.957	0.931	0.944
Discharge letters	RobBERT	0.953	0.974	0.963
Discharge letters	Voting ensemble	0.963	0.966	0.964
General Practitioner entries	Rule-based	0.674	0.801	0.732
General Practitioner entries	BiLSTM	0.889	0.889	0.889
General Practitioner entries	RobBERT	0.950	0.912	0.931
General Practitioner entries	Voting ensemble	0.930	0.886	0.908
Radiology reports	Rule-based	0.901	0.966	0.932
Radiology reports	BiLSTM	0.933	0.934	0.933
Radiology reports	RobBERT	0.960	0.963	0.961
Radiology reports	Voting ensemble	0.955	0.965	0.960
Specialist letters	Rule-based	0.807	0.840	0.823
Specialist letters	BiLSTM	0.922	0.835	0.876
Specialist letters	RobBERT	0.934	0.890	0.911
Specialist letters	Voting ensemble	0.937	0.862	0.898
All letters	Rule-based	0.825	0.892	0.857
All letters	BiLSTM	0.926	0.901	0.913
All letters	RobBERT	0.951	0.938	0.944
All letters	Voting ensemble	0.948	0.924	0.936

ISSN: 1471-2105