Model details

DaBERT-BlameDetector, Also known as BlameBERT, is a zero-shot binary classifier of blame in the Danish Folketing.

It was trained on a custom dataset [runetrust/blame-folketinget-dk][https://huggingface.co/datasets/runetrust/blame-folketinget-dk] and tested on a held out manually annotated test set for a final macro F1 score of 0.80.

Being a Multilingual ModernBERT based model, it should theoretically generalize to other languages, but performance has not been tested.