Identification of Medication Tweets Using Domain-specific Pre-trained Language Models

AbstractIn this paper, we present our approach for task1 of SMM4H 2020. This task involves automatic classification of tweets mentioning medication or dietary supplements. For this task, we experiment with pre-trained models like Biomedical RoBERTa, Clinical BERT and Biomedical BERT. Our approach achieves F1-score of 73.56%.