The Detection of Adverse Drug Reactions in Clinical Text Data Using Transformer Models

Advancements in pharmaceutical science have incorporated various technologies to prevent and revolutionize the treatment of diseases varying from mild ones to fatal cases. One such break with the past is in the identification of drug-related events and actions. This paper aims to harvest information that can be employed to ascertain if a medication is leading to adverse effects in its consumers or patients. Utilized NLP methods that transform text data or corpus into formats that are understandable by machines and hence the machines can be trained to find any inimical events in people due to the medicinal drug consumption. In this paper, the intention is to classify the texts into ADR and non-ADR based on adverse events in each text. The paper has implemented four models - BERT, DistilBERT, BioBERT, and RoBERTa to analyze how transformer models work differently in the given problem statement. Thus, comparing with the above-mentioned methods it can be concluded that RoBERTa outperforms all other techniques.