Wojood - Nested/Flat Arabic NER Models

Wojood is a corpus for Arabic nested Named Entity Recognition (NER). Nested entities occur when one entity mention is embedded inside another entity mention. 550K tokens (MSA and dialect) This repo contains the source-code to train Wojood nested NER.

Online Demo You can try our model using the demo link below

https://sina.birzeit.edu/wojood/

https://arxiv.org/abs/2205.09651

https://huggingface.co/aubmindlab/bert-base-arabertv2/tree/main

Models

• Nested NER (main branch), with micro-F1 score of 0.909551
• Flat NER (flat branch), with micro-F1 score 0.883847

Google Colab Notebooks

You can test our model using our Google Colab notebooks

• Train flat NER: https://gist.github.com/mohammedkhalilia/72c3261734d7715094089bdf4de74b4a
• Evaluate your model using flat NER model: https://gist.github.com/mohammedkhalilia/c807eb1ccb15416b187c32a362001665
• Train nested NER: https://gist.github.com/mohammedkhalilia/a4d83d4e43682d1efcdf299d41beb3da
• Evaluate your data using nested NER model: https://gist.github.com/mohammedkhalilia/9134510aa2684464f57de7934c97138b