Table 8.

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

System	TrIP			TrWP			TrCP			TrAP			TrNAP
	R	P	F	R	P	F	R	P	F	R	P	F	R	P	F
Segment LSTM mean	0.298	0.484	0.369	0.000	NaN	NaN	0.408	0.618	0.491	0.781	0.706	0.742	0.194	0.425	0.266
Sentence LSTM mean	0.268	0.445	0.334	0.056	0.211	0.088	0.464	0.536	0.498	0.744	0.700	0.722	0.209	0.412	0.278
Segment LSTM max	0.263	0.491	0.342	0.049	0.194	0.078	0.403	0.653	0.499	0.776	0.691	0.731	0.178	0.596	0.274
Sentence LSTM max	0.338	0.432	0.380	0.070	0.185	0.102	0.462	0.461	0.461	0.748	0.721	0.735	0.246	0.346	0.287
Roberts et al.	0.298	0.562	0.389	0.035	0.278	0.062	0.565	0.542	0.554	0.814	0.707	0.757	0.199	0.432	0.272
deBruijn et al.	0.177	0.833	0.292	0.000	NaN	NaN	0.327	0.747	0.455	0.730	0.748	0.739	0.126	0.774	0.216
Grouin et al.	0.414	0.458	0.435	0.168	0.774	0.276	0.435	0.550	0.486	0.760	0.676	0.715	0.251	0.495	0.333
Patrick et al.	0.157	0.861	0.265	0.028	0.800	0.054	0.480	0.495	0.487	0.725	0.699	0.712	0.131	0.556	0.212
Jonnalagadda et al.	0.207	0.612	0.309	0.007	0.200	0.014	0.457	0.537	0.494	0.835	0.589	0.691	0.147	0.400	0.215
Divita et al.	0.197	0.780	0.315	0.035	0.833	0.067	0.367	0.715	0.485	0.719	0.702	0.710	0.105	0.690	0.182
Solt et al.	0.313	0.591	0.409	0.056	0.667	0.103	0.493	0.389	0.435	0.743	0.685	0.713	0.220	0.316	0.259
Demner-Fushman et al.	0.369	0.635	0.467	0.126	0.346	0.185	0.491	0.536	0.512	0.712	0.675	0.693	0.199	0.376	0.260
Anick et al.	0.237	0.528	0.328	0.014	0.182	0.026	0.561	0.442	0.495	0.731	0.632	0.678	0.157	0.517	0.241
Cohen et al.	0.096	0.576	0.165	0.007	0.200	0.014	0.356	0.608	0.449	0.729	0.608	0.663	0.052	0.435	0.094