Intent Classification on Arabic-script Darja
87.38AccuracyDziriBERT
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DziriBERTFine-tuning protocol=fine-tuned2026.02 | 87.38 | 0.873 | 0.852 | |
| Rasa NLUArchitecture=DIET2026.02 | 86.98 | 0.87 | 0.845 | |
| AraBERT v2Fine-tuning protocol=fine-tuned2026.02 | 83.5 | 0.831 | 0.807 | |
| TF-IDF + Logistic Regression2026.02 | 79.17 | 0.796 | 0.762 | |
| E5 + MLPBackbone=E52026.02 | 77.7 | 0.776 | 0.732 | |
| E5 + SVMBackbone=E52026.02 | 77.57 | 0.78 | 0.763 | |
| E5 + Random ForestBackbone=E52026.02 | 75.61 | 0.749 | 0.712 | |
| E5 + Logistic RegressionBackbone=E52026.02 | 65.69 | 0.658 | 0.627 |