Using the wisdom of the crowds to find critical errors in biomedical ontologies: a study of SNOMED CT
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
OBJECTIVES: The verification of biomedical ontologies is an arduous process that typically involves peer review by subject-matter experts. This work evaluated the ability of crowdsourcing methods to detect errors in SNOMED CT (Systematized Nomenclature of Medicine Clinical Terms) and to address the challenges of scalable ontology verification. METHODS: We developed a methodology to crowdsource ontology verification that uses micro-tasking combined with a Bayesian classifier. We then conducted a prospective study in which both the crowd and domain experts verified a subset of SNOMED CT comprising 200 taxonomic relationships. RESULTS: The crowd identified errors as well as any single expert at about one-quarter of the cost. The inter-rater agreement (κ) between the crowd and the experts was 0.58; the inter-rater agreement between experts themselves was 0.59, suggesting that the crowd is nearly indistinguishable from any one expert. Furthermore, the crowd identified 39 previously undiscovered, critical errors in SNOMED CT (eg, 'septic shock is a soft-tissue infection'). DISCUSSION: The results show that the crowd can indeed identify errors in SNOMED CT that experts also find, and the results suggest that our method will likely perform well on similar ontologies. The crowd may be particularly useful in situations where an expert is unavailable, budget is limited, or an ontology is too large for manual error checking. Finally, our results suggest that the online anonymous crowd could successfully complete other domain-specific tasks. CONCLUSIONS: We have demonstrated that the crowd can address the challenges of scalable ontology verification, completing not only intuitive, common-sense tasks, but also expert-level, knowledge-intensive tasks.
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.
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 | 0.003 | 0.021 |
| 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 | 0.001 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| 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 it