Adverse Drug Reaction Detection from Social Media Review Using BERT Technology

Abstract

The effects of Adverse Drug Reactions (ADRs) are very harmful to human life; sometimes they cause death. So, the detection of ADRs is very essential. The electronic ADR reporting system has increased, and it is more effective in comparison to the clinical ADR reporting system as patients directly share their opinions on various social media platforms. In the clinical ADR reporting system, it is very difficult to report all the ADRs. ADR extraction from the opinions of patients is very much needed. Advanced Natural Language Processing (NLP) is then essential to apply for the deep learning process. In this article, Bidirectional Encoder Representation from Transformer (BERT) is presented in detection of ADRs from the reviews over the condition on which the drug has been applied. The direction of this work is different from other research work that has been done till now. The condition of the drug to which it has been applied is grouped together to extract the ADRs. This work also helps to develop a drug recommendation system. The f1-score and the accuracy found in the work are very promising, and they outperform the other state-of-the-art machine learning models. This work achieved both a f1-score and accuracy of 0.90 in detecting the ADRs from the analysis of reviews.