Monitoring Opioid-Related Social Media Chatter Using Natural Language Processing and Large Language Models: Temporal Analysis
Bibliographic record
Abstract
Background: Opioid overdose is a global public health emergency, with the United States experiencing high rates of morbidity and mortality due to prescription and illicit opioid use. Traditional public health monitoring systems often fail to provide real-time insights, limiting their capacity for early detection and intervention. Social media platforms, especially Reddit, offer a promising alternative for timely toxicovigilance due to the abundance of user-generated, real-time content. Objective: This study aimed to explore the use of Reddit as a real-time, high-volume source for toxicovigilance and develop an automated system that can classify and analyze opioid-related social media posts to detect behavioral patterns and monitor the evolution of public discourse on opioid use. Methods: To investigate evolving social media discourse around opioid use, we collected a large-scale dataset from Reddit spanning 6 years, from January 1, 2018, to December 30, 2023. Using a comprehensive opioid lexicon-including formal drug names, street slang, common misspellings, and abbreviations-we filtered relevant posts for further analysis. A subset of these data was manually annotated according to well-defined annotation guidelines into 4 categories: self-misuse, external misuse, information, and unrelated, with distributions of 37.21%, 27.25%, 27.57%, and 7.97%, respectively. To automate the classification of opioid-related chatter, we developed a robust natural language processing pipeline leveraging classical machine learning algorithms, deep learning models, and transformer-based architecture, and fine-tuned a state-of-the-art large language model (LLM; OpenAI GPT-3.5 Turbo). In the final stage, the trained LLM was deployed on an unlabeled dataset comprising 74,975 additional Reddit chatter posts. This enabled a detailed temporal analysis of opioid-related discussions, aligned with 6 years of opioid-related death records from the Centers for Disease Control and Prevention (CDC). For this study, self-misuse and external misuse were merged into a misuse category for direct comparison with the CDC's mortality data, examining whether trends in social media discourse on opioid misuse reflect patterns in real-world mortality statistics. Results: The fine-tuned GPT-3.5 Turbo model achieved the highest classification accuracy of 0.93, outperforming the baseline (random forest 0.85) by representing a performance improvement of 9.14% over the machine learning model. The temporal analysis of the unlabeled data revealed evolving trends in opioid-related discussions, indicating shifts in user behavior and overdose-related chatter over time. To quantify this relationship, we calculated the Pearson correlation coefficient between misuse-related posts and CDC death records (r=0.854). This correlation was statistically significant (P<.001), indicating a strong positive relationship between web-based discussions and CDC mortality data. Conclusions: This study demonstrates the potential of integrating advanced natural language processing techniques and LLMs with social media data to support real-time public health surveillance. Reddit provides a valuable platform for identifying emerging trends in opioid use and overdose risk. The proposed system offers a proactive tool for researchers, clinicians, and policymakers to better understand and respond to the opioid crisis.
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 machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.003 | 0.010 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.004 | 0.002 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.001 | 0.002 |
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 source (direct Gemma or distilled Codex), 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".