Cross-sector generalization of accident-process role classification in occupational accident narratives
Task-adapted models classify French accident narratives across sectors at about 86% balanced accuracy.
Researchers evaluate cross-sector generalization of role classification on French occupational accident narratives. Classifiers trained only on 42,244 factual units from 6,040 construction narratives label work situation, unfavourable condition, accident event, and consequence. Without retraining, leading task-adapted models averaged 85.6–85.8% balanced accuracy on metallurgy, chemistry–plastics, and a company corpus. Task-specific adaptation outperformed frozen pretrained representations.
- Role classifiers trained only on 6,040 construction-sector narratives
- 42,244 factual units labeled into four accident-process roles
- Tested on metallurgy, chemistry-plastics, and a company corpus
- Best task-adapted runs averaged 85.6% to 85.8% balanced accuracy
From the original article
1 min read at arxiv.orgOccupational accident narratives contain valuable information about work situations, unfavourable conditions, accident events, and their consequences. Automatically structuring these narratives can facilitate large-scale accident analysis and support occupational risk prevention.
By Aho Yapi · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.