Detecting Substance Use Disorder Using Social Media Data and the Dark Web: Time- and Knowledge-Aware Study

Area of Science:

• Public Health
• Computational Social Science
• Data Science

Background:

• The opioid crisis is a significant public health issue in the United States.
• Understanding user perceptions of synthetic opioids is crucial for developing effective interventions.
• Limited evidence exists on the direct relationship between substance misuse and mental health, impacting treatment accessibility.

Purpose of the Study:

• To analyze social media posts concerning substance use and opioids sold on cryptomarkets.
• To apply deep learning models to gauge user sentiment and emotions regarding various synthetic opioids.
• To identify correlations between specific drugs and user emotional responses, including fear, sorrow, and optimism.

Main Methods:

• Utilized a drug abuse ontology and advanced deep learning models, including Bidirectional Encoder Representations From Transformers (BERT).
• Crawled cryptomarket data, extracting posts related to fentanyl, its analogs, and novel synthetic opioids.
• Performed topic analysis on sentiments and emotions, correlating them with drug-related topics and employing time-aware neural models.

Main Results:

• The most effective deep learning model achieved a macro-F1-score of 82.12 and recall of 83.58 in identifying substance use disorder.
• Identified distinct sentiment and emotional responses associated with different synthetic opioids.
• Correlated user responses with topics such as pain relief, addiction, and withdrawal symptoms.

Conclusions:

• Provides valuable insights into public perception of synthetic opioids through social media sentiment analysis.
• Findings can inform public health policies and interventions to mitigate substance misuse and the opioid crisis.
• Demonstrates the efficacy of deep learning for analyzing social media data in public health research.