Automated Risk Assessment of Opioid Use: Analysis Using Pre-Trained Transformers on Social Media Data

Area of Science:

• Computational linguistics
• Public health informatics
• Machine learning for healthcare

Background:

• The opioid epidemic causes significant mortality and addiction globally.
• Automated tools are needed for faster overdose detection and risk assessment.
• Social media platforms like Reddit offer valuable self-reported data on opioid misuse.

Purpose of the Study:

• To develop an automated system for detecting opioid overdose risks.
• To classify substances as high-risk or low-risk using social media posts.
• To enhance early intervention and harm reduction strategies.

Main Methods:

• Constructed and manually annotated a novel dataset from Reddit posts.
• Utilized a BioBERT (Bidirectional Encoder Representations from Transformers for Biomedical text mining) model enhanced with a custom attention mechanism.
• Evaluated performance using 5-fold cross-validation and compared against baseline models, including XGBoost (Extreme Gradient Boosting).

Main Results:

• The BioBERT model with custom attention achieved a 0.99 F1-score, surpassing the best baseline (XGBoost at 0.97).
• A paired t-test confirmed a statistically significant performance improvement (P=.003).
• The model demonstrated robust and accurate overdose risk detection capabilities.

Conclusions:

• Social media data combined with advanced NLP models can create effective opioid overdose detection systems.
• The BioBERT model with custom attention offers state-of-the-art performance for real-time intervention.
• This technology supports timely harm reduction efforts in the opioid crisis.