A Weak Supervision Approach for Monitoring Recreational Drug Use Effects in Social Media
arXiv Preprint Archive September 18, 2025 Lucía Prieto-Santamaría, Alba Cortés Iglesias, Claudio Vidal Giné et al.
Social media posts on Twitter can reveal how people describe the effects of recreational drugs like ecstasy, GHB, and 2C-B. By analyzing over 92,000 tweets using slang terms and biomedical concept extraction, researchers identified whether each post reported a positive or negative effect. Machine learning classifiers, particularly eXtreme Gradient Boosting with cost-sensitive learning, predicted tweet polarity with high accuracy (F1 = 0.885, AUPRC = 0.934). The findings suggest that Twitter data can detect substance-specific effects and support real-time drug monitoring and characterization of effects.