TABLE 3.

CASBERT performance over three query-entity pair sets and seven searching methods compared to the bag-of-words method (BM25) measured using mAP@10 and mRR. The numbers of entities in the PMR and the BioModels are 4,652 and 54,456 respectively.

Method	noPredicate		withPredicate		Combine
	mAP@10	mRR	mAP@10	mRR	mAP@10	mRR
PMR-CA
macro	0.645216	0.643424	0.572291	0.597649	0.601406	0.602604
macroWP	0.626565	0.624992	0.585014	0.606947	0.604012	0.608011
micro	0.619483	0.617890	0.534427	0.549304	0.592228	0.595516
microWP	0.606255	0.608560	0.504055	0.515426	0.574869	0.579472
mixed	0.656636	0.656326	0.572109	0.583003	0.623287	0.624638
mixedWP	0.641843	0.641940	0.585494	0.612151	0.615885	0.621419
mixedCl	0.660346	0.659399	0.589960	0.615694	0.626420	0.631739
BM25	0.459375	0.443034	0.464496	0.487373	0.456505	0.457420
BioModels-CA
macro	0.345486	0.340106	0.310285	0.302129	0.356034	0.347905
macroWP	0.330100	0.326029	0.326831	0.325134	0.365903	0.362892
micro	0.294353	0.291914	0.283026	0.283291	0.312949	0.309471
microWP	0.301564	0.296179	0.274491	0.271451	0.314506	0.309211
mixed	0.347681	0.345045	0.326664	0.323393	0.370148	0.365433
mixedWP	0.335051	0.331222	0.344296	0.344932	0.379058	0.377573
mixedCl	0.348054	0.344794	0.343567	0.344058	0.379351	0.377557
BM25	0.240671	0.231515	0.294758	0.294779	0.280370	0.277879

The values in bold are maximum performance measured on three sets of query-entity pairs using mAP@10 and mRR values.