Sentiment Analysis Using a Large Language Model–Based Approach to Detect Opioids Mixed With Other Substances Via Social Media: Method Development and Validation
Bibliographic record
Abstract
Background: The opioid crisis poses a significant health challenge in the United States, with increasing overdoses and death rates due to opioids mixed with other illicit substances. Various strategies have been developed by federal and local governments and health organizations to address this crisis. One of the most significant objectives is to understand the epidemic through better health surveillance, and machine learning techniques can support this by identifying opioid users at risk of overdose through the analysis of social media data, as many individuals may avoid direct testing but still share their experiences online. Objective: In this study, we take advantage of recent developments in machine learning that allow for insights into patterns of opioid use and potential risk factors in a less invasive manner using self-reported information available on social platforms. Methods: This study used YouTube comments posted between December 2020 and March 2024, in which individuals shared their self-reported experiences of opioid drugs mixed with other substances. We manually annotated our dataset into multiclass categories, capturing both the positive effects of opioid use, such as pain relief, euphoria, and relaxation, and negative experiences, including nausea, sadness, and respiratory depression, to provide a comprehensive understanding of the multifaceted impact of opioids. By analyzing this sentiment, we used 4 state-of-the-art machine learning models, 2 deep learning models, 3 transformer models, and 1 large language model (GPT-3.5 Turbo) to predict overdose risks to improve health care response and intervention strategies. Results: Our proposed methodology (GPT-3.5 Turbo) was highly precise and accurate, helping to automatically identify sentiment based on the adverse effects of opioid drug combinations and high-risk drug use in YouTube comments. Our proposed methodology demonstrated the highest achievable F1-score of 0.95 and a 3.26% performance improvement over traditional machine learning models such as extreme gradient boosting, which demonstrated an F1-score of 0.92. Conclusions: This study demonstrates the potential of leveraging machine learning and large language models, such as GPT-3.5 Turbo, to analyze public sentiment surrounding opioid use and its associated risks. By using YouTube comments as a rich source of self-reported data, the study provides valuable insights into both the positive and negative effects of opioids, particularly when mixed with other substances. The proposed methodology significantly outperformed traditional models, contributing to more accurate predictions of overdose risks and enhancing health care responses 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 | 0.004 | 0.009 |
| Meta-epidemiology (narrow) | 0.002 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.003 | 0.003 |
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".