dbmdz/flair-historic-ner-onb

We use BPEmbeddings instead of the combination of Wikipedia, Common Crawl and character embeddings (as used in the paper) to save space and training/inferencing time . Paper reported an averaged F1-score of 85.31.31%. Paper reported average F1 score of 85-85.69.75% for model uploads . Paper published an average score of 86.69% for new model trained on the ONB dataset . Paper: Robust Named Named Entity Recognition for Historic German German. [Towards Robust named entity Recognition. For Historic German.org.uk: We release a new model based on our paper. The,