UMCU/DutchMedicalTextDetector_v1_HEADONLY

We used GPT4.1-nano to classify generic texts from OSCAR as non-medical/medical using PubScience. We labeled 400.000 texts, with about 40.000 labeled as positive. We then trained a SequenceClassifier on 80.000 samples with a 50/50 class ratio.

This can be used e.g. to approximately identify medical texts in general corpora.