Extracting circumstances of Covid-19 transmission from free text with large language models
Bibliographic record
Abstract
Identifying the circumstances of transmission of an emerging infectious disease rapidly is central for mitigation efforts. Here, we explore how large language models (LLMs) can automatically extract such circumstances from free-text descriptions in online surveys, in the context of Covid-19. In a nationwide study conducted online in France, we enrolled 545,958 adults with recent SARS-CoV-2 infection and inquired about the circumstances of transmission in both closed-ended and open-ended questions. First, we trained a classification model based on a pretrained LLM to predict one of seven predefined infection contexts (Work, Family, Friends, Sports, Cultural, Religious, Other) from the free text in answers to open-ended questions. We achieved an unbalanced accuracy of 75%, which increased to 91% when eliminating the 43% highest entropy responses. Second, we used topic modeling to define clusters of transmission circumstances agnostically. This led to 23 clusters, which agreed with the seven predefined infection contexts, but also provided finer details on previously undefined circumstances of transmission. Our study suggests that LLM-based analysis of free text may alleviate the need for closed-ended questions in epidemiological surveys and enable insights into previously unsuspected circumstances of transmission. This approach is poised to accelerate and enrich the acquisition of epidemiological insights in future pandemics. Open-ended survey questions may provide useful detail on possible venues of transmission of infectious diseases, but data are difficult to analyse at scale. Here, the authors use large language models to extract potential transmission venues in ~80,000 responses to an open-ended COVID-19 survey question in France.
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 | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.003 | 0.000 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 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".