. Author manuscript; available in PMC: 2024 Apr 24.

Published in final edited form as: J Biomed Inform. 2022 Mar 26;129:104059. doi: 10.1016/j.jbi.2022.104059

Table 2.

Performance of the re-ranking model without the selection embedding component

Method	Precision	Recall	F1-score	Precision	Recall	F1-score
Method	BiolarkGSC+			COPD-HPO
MMRerank w/o SE	0.605(−20%)	0.460(−23%)	0.523(−22%)	0.463(−44%)	0.615(−21%)	0.528(−34%)
MNIRerank w/o SE	0.567(−28%)	0.483(−20%)	0.522(−24%)	0.445(−45%)	0.564(−23%)	0.497(−35%)
PTRerank w/o SE	0.753(−11%)	0.407(−43%)	0.528(−31%)	0.683(−18%)	0.506(−34%)	0.581(−28%)
	BiolarkGSC+ Testing Set			COPD-HPO Testing Set
MMRerank w/o SE	0.581(−27%)	0.377(−7%)	0.457(−15%)	0.478(−33%)	0.623(−4%)	0.541(−20%)
MNIRerank w/o SE	0.714(−14%)	0.349(−21%)	0.468(−19%)	0.658(−1%)	0.590(−1%)	0.622(−1%)
PTRerank w/o SE	0.580(−19%)	0.218(−50%)	0.317(−41%)	0.637(−10%)	0.558(−21%)	0.595(−15%)

Note. w/o SE represents without the selection embedding component. Each percentage number in the brackets is the change compared with the sampe approach with selection embedding.