Identifying Self-Disclosures of Use, Misuse and Addiction in Community-based Social Media Posts.

Area of Science:

• Public Health
• Computer Science
• Social Sciences

Background:

• Opioid overdose is a national public health emergency in the US, causing over 500,000 deaths.
• Identifying patients at risk for opioid use disorder (OUD) requires effective tools for medical practitioners.
• Social media platforms like Reddit offer insights into sensitive drug-related behaviors through user self-disclosure.

Purpose of the Study:

• To develop and evaluate models for classifying six phases of opioid use (Medical Use, Misuse, Addiction, Recovery, Relapse, Not Using) from Reddit posts.
• To investigate the impact of span-level extractive explanations on annotation quality and model performance in OUD phase classification.
• To assess the effectiveness of state-of-the-art models in supervised, few-shot, and zero-shot learning settings for OUD continuum analysis.

Main Methods:

• Collected and annotated a corpus of 2500 opioid-related Reddit posts, labeling six distinct phases of opioid use.
• Annotated span-level extractive explanations for each post to enhance understanding and model training.
• Evaluated various state-of-the-art machine learning models using supervised, few-shot, and zero-shot learning approaches.

Main Results:

• Classifying the phases of opioid use disorder is context-dependent and presents significant challenges.
• The inclusion of explanations during model development led to a substantial increase in classification accuracy.
• Error analysis highlighted the complexity and contextual nature of identifying OUD phases.

Conclusions:

• Span-level explanations are beneficial for improving model performance in high-stakes domains like OUD research.
• Machine learning models, especially when augmented with explanations, show promise for analyzing the opioid use disorder continuum.
• This approach can aid in developing better tools for identifying at-risk patients and understanding OUD progression.