10.1101/2020.05.20.107003

On the effectiveness of small, discriminatively pre-trained language representation models for biomedical text mining