Table 10.

Class-wise performance of LSTM models with medical word embedding on TeP relations in comparison with challenge participating systems. The best F-measure for each relation is in bold.

System	TeRP			TeCP
	R	P	F	R	P	F
Segment LSTM mean	0.831	0.858	0.844	0.430	0.680	0.527
Sentence LSTM mean	0.828	0.806	0.817	0.396	0.679	0.501
Segment LSTM max	0.830	0.869	0.849	0.427	0.719	0.536
Sentence LSTM max	0.816	0.833	0.824	0.456	0.549	0.498
Roberts et al.	0.906	0.825	0.864	0.456	0.594	0.516
deBruijn et al.	0.880	0.842	0.861	0.316	0.857	0.462
Grouin et al.	0.881	0.813	0.846	0.391	0.612	0.477
Patrick et al.	0.840	0.840	0.840	0.430	0.614	0.506
Jonnalagadda et al.	0.911	0.784	0.843	0.400	0.596	0.479
Divita et al.	0.886	0.793	0.837	0.245	0.818	0.377
Solt et al.	0.826	0.842	0.834	0.536	0.577	0.556
Demner-Fushman et al.	0.733	0.872	0.796	0.393	0.594	0.473
Anick et al.	0.848	0.765	0.804	0.475	0.597	0.529
Cohen et al.	0.861	0.766	0.810	0.369	0.599	0.457