Table 3.

Evaluation results of the named entity recognition models on the test set (lenient evaluation).

Entity	RIWEa				PWEb				(PWE+CE)c				(PWE+SFE)d
	Precision	Recall	F-score	Precision		Recall	F-score	Precision		Recall	F-score	Precision		Recall	F-score
Drug	0.942	0.892	0.917	0.963		0.930	0.946	0.946		0.953	0.949	0.952		0.947	0.950e
Strength	0.977	0.959	0.968	0.979		0.970	0.975	0.973		0.976	0.974	0.977		0.977	0.977
Duration	0.893	0.706	0.789	0.883		0.762	0.818	0.910		0.698	0.790	0.903		0.786	0.840
Route	0.964	0.928	0.946	0.964		0.938	0.951	0.956		0.948	0.952	0.953		0.943	0.948
Form	0.964	0.935	0.949	0.965		0.940	0.952	0.969		0.944	0.956	0.972		0.932	0.951
Dosage	0.928	0.912	0.920	0.932		0.931	0.931	0.931		0.928	0.929	0.928		0.931	0.930
Frequency	0.945	0.925	0.935	0.965		0.952	0.959	0.980		0.933	0.956	0.968		0.968	0.968
Reason	0.771	0.458	0.575	0.821		0.497	0.620	0.860		0.452	0.593	0.621		0.653	0.637
Micro	0.943	0.863	0.901	0.951		0.892	0.921	0.950		0.894	0.921	0.927		0.913	0.920
Macro	0.936	0.840	0.883	0.951		0.876	0.910	0.949		0.884	0.914	0.923		0.901	0.910

^aRIWE: bidirectional long-short term memory with conditional random fields with random word embeddings.

^bPWE: bidirectional long-short term memory with conditional random fields with pretrained word embeddings.

^c(PWE+CE): bidirectional long-short term memory with conditional random fields with pretrained word embeddings and character embeddings.

^d(PWE+SFE): bidirectional long-short term memory with conditional random fields with pretrained word embeddings and semantic-feature embeddings.

^eThe best results for each metric are italicized.