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

Table 7

The comparison of overall performance among MIL-dEM-SL, MIL-dEM-T, advanced machine learning methods, and MIL-iEM-T using fivefold cross-validation.

Models	BOW				S–P
	P	R	F1	Acc.	P	R	F1	Acc.
Supervised learning
MIL-dEM-TF-S-SL^1	—	—	—	—	0.904	0.993	0.946	0.939
MISVM-TFIDF^2	0.918	0.885	0.901	0.895	0.799	0.733	0.765	0.735
MINB-B	0.864	0.896	0.880	0.867	0.619	0.701	0.744	0.691
MILR-B^3	0.869	0.852	0.861	0.850	0.718	0.783	0.749	0.692
Transductive learning
MIL-dEM-B-S-T	—	—	—	—	0.934	0.975	0.954	0.949
TSVM-B	0.898	0.881	0.889	0.881	0.873	0.865	0.869	0.859
MIL-iEM-TF-S-T	0.749	0.850	0.797	0.764	0.844	0.838	0.841	0.827

^1,4γ = , β = , α = . ²Polynomial kernel, C = 10. ³Collective MI assumption, geometric mean for posteriors.