FactCheckEC-Sci: detección de afirmaciones científicas en un corpus ecuatoriano de fact-checking en español

Resumen

This paper presents FactCheckEC-Sci, a resource for detecting scientific claims in an Ecuadorian Spanish fact-checking corpus. From 6299 records, 1804 candidate posts were filtered and manually reviewed, yielding 419 instances of scientific content. The annotation scheme considers three non-exclusive categories: verifiable scientific claim, scientific reference, and research context. Transformer-based models and a TF-IDF and Logistic Regression baseline were evaluated; among them, BETO achieved the best performance. In addition, automatic scholarly evidence retrieval was explored as support for verification. The results show the potential of the corpus to strengthen fact-checking tools in Spanish.