Table 4.

Entity-level performance of PubMedBERT on the gold test set. Original scores refer to the performance of the model before self-training; self-training scores refer to the performance of the model at the best iteration (6th iteration) of self-training. ‘R’ and ‘P’ refer to recall and precision, respectively

	Original scores			Self-training scores
	F1	R	P	F1	R	P
Comparator	16.0	10.0	40.0	48.5	40.0	61.5
Induction	49.1	50.6	47.7	48.0	49.4	46.6
Intervention	70.2	76.1	65.2	69.8	74.6	65.6
Outcome	65.4	70.6	60.9	66.9	70.6	63.6
Species	98.1	100.0	96.4	98.1	100.0	96.4
Strain	63.4	72.2	56.5	70.0	77.8	63.6
Overall	69.9	73.4	66.7	71.0	74.0	68.2