Intentional overdose (OD) of over-the-counter (OTC) and prescription drugs is becoming a significant social issue all over the world. While previous research has focused on drug misuse, there has been limited analysis using social networking service data. This study aims to analyze posts related to a drug overdose on Twitter® (X®) to understand the characteristics and trends of drug misuse, and to examine the applicability of social media in understanding the current situation of OD through natural language processing techniques. We collected posts in Japanese containing the term "OD" from January 10 to February 8, 2023, and analyzed 30203 posts. Using a pre-trained, fine-tuned bidirectional encoder representations from transformers (BERT) model, we classified the posts into categories, including direct mentions of OD. We examined the content for drug types and emotional context. Among the 5283 posts categorized as "Posts describing ODing," about one-third included specific drug names or related terms. The most frequently mentioned OTC drugs included active ingredients such as codeine, dextromethorphan, ephedrine, and diphenhydramine. Prescription drugs, particularly benzodiazepines and pregabalin, were also common. Tweets peaked at midnight, suggesting a link between negative emotions and potential OD incidents. Our classifier showed high accuracy in distinguishing OD-related posts. Analyzing Twitter® posts provides valuable insights into the patterns and emotional contexts of drug misuse. Monitoring social networking services for OD-related content could help identify high-risk individuals and inform prevention strategies. Enhanced monitoring and public awareness are crucial to reducing the risks associated with both OTC and prescription drug misuse.
Keywords: natural language processing; overdose; social media.