Adverse Drug Reaction Posts Detection with a Bi-LSTM based approach

Social Network Services (SNS) is currently the most active communication platform, generating big data every day. In the field of pharmacovigilance, it is also an interesting research topic to extract meaningful information from accumulated data from SNS data. In this work, we propose a Recurrent Neural Network (RNN) based classification model to extract Adver Drug Reactions (ADR) posts from social network service (SNS) data. For Ketoprofen drugs with high prescription frequency and high number of posts, posts from Naver Blog and Cafe (2005-2020) were secured, and the final 3,828 cases were analyzed. As a result, three types of lexicons (drugs name, ADR, and stop words) were defined for Ketoprofen, and based on this, 87% accuracy was obtained based on the Bi-LSTM classification model. Extra drug was also verified through the entire process above, with an accuracy of 80%. It is expected to provide convenience in extracting ADR posts through the developed Bi-LSTM classification model.