Large-Scale Deep Learning–Enabled Infodemiological Analysis of Substance Use Patterns on Social Media: Insights From the COVID-19 Pandemic
Bibliographic record
Abstract
BACKGROUND: The COVID-19 pandemic intensified the challenges associated with mental health and substance use (SU), with societal and economic upheavals leading to heightened stress and increased reliance on drugs as a coping mechanism. Centers for Disease Control and Prevention data from June 2020 showed that 13% of Americans used substances more frequently due to pandemic-related stress, accompanied by an 18% rise in drug overdoses early in the year. Simultaneously, a significant increase in social media engagement provided unique insights into these trends. Our study analyzed social media data from January 2019 to December 2021 to identify changes in SU patterns across the pandemic timeline, aiming to inform effective public health interventions. OBJECTIVE: This study aims to analyze SU from large-scale social media data during the COVID-19 pandemic, including the prepandemic and postpandemic periods as baseline and consequence periods. The objective was to examine the patterns related to a broader spectrum of drug types with underlying themes, aiming to provide a more comprehensive understanding of SU trends during the COVID-19 pandemic. METHODS: We leveraged a deep learning model, Robustly Optimized Bidirectional Encoder Representations from Transformers Pretraining Approach (RoBERTa), to analyze 1.13 billion Twitter (subsequently rebranded X) posts from January 2019 to December 2021, aiming to identify SU posts. The model's performance was enhanced by a human-in-the-loop strategy that subsequently enriched the annotated data used during the fine-tuning phase. To gain insights into SU trends over the study period, we applied a range of statistical techniques, including trend analysis, k-means clustering, topic modeling, and thematic analysis. In addition, we integrated the system into a real-time application designed for monitoring and preventing SU within specific geographic locations. RESULTS: Our research identified 9 million SU posts in the studied period. Compared to 2019 and 2021, the most substantial display of SU-related posts occurred in 2020, with a sharp 21% increase within 3 days of the global COVID-19 pandemic declaration. Alcohol and cannabinoids remained the most discussed substances throughout the research period. The pandemic particularly influenced the rise in nonillicit substances, such as alcohol, prescription medication, and cannabinoids. In addition, thematic analysis highlighted COVID-19, mental health, and economic stress as the leading issues that contributed to the influx of substance-related posts during the study period. CONCLUSIONS: This study demonstrates the potential of leveraging social media data for real-time detection of SU trends during global crises. By uncovering how factors such as mental health and economic stress drive SU spikes, particularly in alcohol and prescription medication, we offer crucial insights for public health strategies. Our approach paves the way for proactive, data-driven interventions that will help mitigate the impact of future crises on vulnerable populations.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.002 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".