Distant Supervision with Transductive Learning for Adverse Drug Reaction Identification from Electronic Medical Records

Table 6

The effectiveness of MIL-dEM-S-SL and MIL-dEM-S-T comparison across three types of initial weight on fivefold cross-validation with soft decision making.

	Models	B			TF			TFIDF
		P	R	F1	P	R	F1	P	R	F1
Supervised learning	MIL-dEM-S-SL^1
	S–P	0.883	0.978	0.928	0.904	0.993	0.946	0.890	0.993	0.938
	L–P	0.896	0.962	0.928	0.898	0.978	0.936	0.889	0.976	0.930
Transductive learning	MIL-dEM-S-T
	S–P	0.934	0.975	0.954	0.901	0.942	0.921	0.881	0.951	0.915
	L–P	0.926	0.962	0.944	0.919	0.916	0.918	0.875	0.945	0.909
	MIL-dEM-S-T
	S–P	0.839	0.907	0.872	0.635	0.925	0.754	0.686	0.916	0.784
	L–P	0.850	0.889	0.869	0.663	0.900	0.763	0.714	0.887	0.791
	MIL-dEM-S-T
	S–P	0.830	0.889	0.859	0.581	0.607	0.594	0.647	0.682	0.664
	L–P	0.843	0.865	0.854	0.597	0.619	0.608	0.657	0.679	0.668

^1,2γ = , β = , α = ; ^3,4γ = , β = , α = . B: binary frequency; TF: term frequency; TFIDF: term frequency-inverse document frequency.