查看文章

In an era where medicine and technology are closely intertwined, sources of patient-generated data such as social media content are being explored to extract important information for the study of public health and patient trajectories. Drug related mentions present in Twitter posts are a particular use case, as the process of automatically extracting drug related mentions from tweets can provide novel relevant information for pharmacoepidemiologic studies. In this paper, we describe the system developed by the BIT. UA team from the University of Aveiro during the participation in BioCreative VII Track 3 on automatic extraction of medication names in tweets. The system consists of an end-to-end deep learning architecture based on transformers, and was used in all three submitted runs for the challenge. Run 1 obtained the best results on strict evaluation (F1-score of 0.6810) whereas Run 3 performed better on overlapping evaluation (F1-score of 0.7700).