| Resumen |
The global opioid crisis, driven significantly by highly addictive synthetic opioids such as heroin and fentanyl, presents severe social, economic, and psychological challenges. While overdose prevention has received considerable attention, the withdrawal phase remains a critical yet underexplored aspect that significantly influences relapse risk. Withdrawal symptoms—including intense cravings, insomnia, muscle aches, and anxiety—often compel individuals to resume opioid use, increasing the likelihood of overdose. Traditional healthcare systems struggle to monitor withdrawal experiences effectively outside clinical settings. Social media platforms like Twitter and Reddit have emerged as vital sources of real-time, user-generated data on withdrawal experiences and help-seeking behaviors. Leveraging this data, natural language processing (NLP) offers promising avenues for automatic detection and classification of withdrawal-related posts to facilitate timely interventions. In this study, we developed a novel dataset annotated for multiclass classification of heroin and fentanyl-related user behavior into withdrawal symptoms, relapse risk, and help-seeking intent. We introduce Wise, a hybrid model combining Logistic Regression and Support Vector Machine, which effectively extracts local patterns from noisy social media text using TF-IDF features. Wise outperforms traditional classifiers, achieving 76% accuracy and surpassing Decision Tree and K-Nearest Neighbors models by 8.57% and 43.40%, respectively. Our findings demonstrate the potential of NLP-driven approaches in enhancing opioid withdrawal monitoring and support systems, offering valuable tools for healthcare providers and policymakers in addressing this public health crisis. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2026. |
