Fine-tuned BERTurk model for extracting entities from Turkish volunteer
activity announcements.
F1 = 0.846 on a stress test set with informal language and OOV vocabulary.
1from transformers import pipeline
2ner = pipeline("token-classification",
3 model="francky-saah/turkish-volunteer-ner",
4 aggregation_strategy="simple")
5result = ner("İstanbul'da hafta sonu çevre temizliği etkinliği")
6print(result)
3000 annotated Turkish sentences (BIO scheme), split 2000/500/500 for
train/val/test.
Part of a project on cache-aware sequential recommendation for Turkish
volunteer activities.
