Hybrid natural language processing tool for semantic annotation of medical texts in Spanish

BMC Bioinformatics

Table 3 Fine-tuning hyperparameters of the tested models

Model	B	Mx Ep	LR	Optim	Pat	Seed
Bi-LSTM-CRF (FLAIR)	16	100	0.1	SGD	5	Random
RoBERTa, EriBERTa,	16	20	2e-05	Adam	5	{100, 200, 300, 400, 500}
mBERT and mDeBERTA vs 3	16	20	2e-05	Adam	5	{100, 200, 300, 400, 500}
CLIN-X-ES	8	30	2e-05	Adam	5	{100, 200, 300, 400, 500}

B: ‘batch’; LR: ‘fine-tune learning rate’; Mx Ep: ‘maximum number of epochs’; Pat: ‘Patience’;
Optim: ‘Optimizer’; SGD: ‘stochastic gradient descent’

ISSN: 1471-2105