alger-ia/dziribert

DziriBERT: a Pre-trained Language Model for the Algerian Dialect # DziriBERT DziriBERT is the first Transformer-based Language Model that has been pre-trained specifically for the Algerian Dialect. It handles Algerian text contents written using both Arabic and Latin characters. It sets new state of the art results on Algerian text classification datasets (see below). The model is publicly available at: huggingface.co. For more information, please visit our paper: arxiv.org **\* New: \*** A fine-tuned version for sentiment classification available here. ## Evaluation Results on the Twifil dataset: | Model | Sentiment acc. | Emotion acc. | | ---------------------------------------- |:--------------:|:------------:| | bert-base-multilingual-cased | 74.2 % | 62.0 % | | aubmindlab/bert-base-arabert | 73.9 % | 64.6 % | | CAMeL-Lab/bert-base-arabic-camelbert-mix | 77.7 % | 69.1 % | | qarib/bert-base-qarib | 78.8 % | 68.9 % | | UBC-NLP/MARBERT | 80.5 % | 70.2 % | | alger-ia/dziribert | 80.5 % | 70.4 % | Results on the Narabizi dataset: | Model | Sentiment acc. | Topic acc. | | ---------------------------------------- |:--------------:|:------------:| | bert-base-multilingual-cased | 52.6 % | 49.3 % | | aubmindlab/bert-base-arabert | 49.1 % | 42.8 % | | CAMeL-Lab/bert-base-arabic-camelbert-mix | 49.4 % | 47.0 % | | qarib/bert-base-qarib | 55.0 % | 45.7 % | | UBC-NLP/MARBERT | 58.0 % | 49.0 % | | alger-ia/dziribert | 63.5 % | 62.8 % | In order to reproduce these results, please install the following requirements: ```bash pip install -r requirements.txt …