arabert-sarcasm-detector

This model is a fine-tuned aubmindlab/bert-base-arabertv02-twitter that was fine-tuned on the ArSarcasT dataset. It achieves the following results on the evaluation set:

Eval Datatset	Accuracy	F1	Precession	Recall
ArSarcasT	0.839	0.730	0.743	0.761
iSarcasmEVAL	0.897	0.633	0.646	0.620
AraSarcasmV2	0.769	0.553	0.587	0.523
IDAT	0.807	0.807	0.775	0.842

Model description

Fine-tuned aubmindlab/bert-base-arabertv02-twitter on Sarcastic tweets dataset for sarcasm detection text classification.

Intended uses & limitations

More information needed

Training and evaluation data

• Training dataset: ArSarcasT development split.
• Evaluation Datasets:
  • ArSarcasm-v2 test dataset.
  • iSarcasmEVAL test dataset.
  • ArSarcasT test dataset.

Training procedure

Fine-tuning, 3 epochs.

Training hyperparameters

The following hyperparameters were used during training:

• optimizer: None
• training_precision: float32

Training results

Framework versions

• Transformers 4.28.1
• TensorFlow 2.12.0
• Tokenizers 0.13.3

Paper Citation

If you use this fine-tuned model based on the original ARABERT model, please cite the following paper:

Galal, M. A., Yousef, A. H., Zayed, H. H., & Medhat, W. (2024). Arabic sarcasm detection: An enhanced fine-tuned language model approach. Ain Shams Engineering Journal, 15(6), 102736. https://doi.org/10.1016/j.asej.2024.102736

Dataset Repo

https://github.com/Mabdelaziz/ArSarcasT