You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

MARBERT Fine-tuned for Darija Intent Classification

A fine-tuned MARBERTv2 model for intent classification in Moroccan Arabic (Darija). Designed for customer service automation and contact center AI applications.

🔗 Related Projects

Resource	Link
🚀 Live API	mohammedmediani-darija-aicc-api.hf.space
📺 Interactive Demo	HuggingFace Space
💻 API Source Code	GitHub Repository

📋 Supported Intents

The model classifies text into 9 customer service intents:

Intent	Description
`consulter_solde`	Check account balance or credit
`reclamer_facture`	Bill complaint or dispute
`declarer_panne`	Report technical issue
`info_forfait`	Request plan information
`recuperer_mot_de_passe`	Password recovery
`salutations`	Greetings
`remerciements`	Thanks
`demander_agent_humain`	Request human agent
`hors_scope`	Out of scope

🚀 Quick Start

from transformers import pipeline

classifier = pipeline("text-classification", model="mediani/marbert-fine-tuned-darija-aicc")

# Test examples
texts = [
    "بغيت نعرف شحال باقي ليا في رصيدي",
    "ماكيخدمش عندي لانترنيت",
    "شكرا بزاف على المساعدة"
]

for text in texts:
    result = classifier(text)[0]
    print(f"'{text}' → {result['label']} ({result['score']:.2%})")

Output:

'بغيت نعرف شحال باقي ليا في رصيدي' → consulter_solde (99.87%)
'ماكيخدمش عندي لانترنيت' → declarer_panne (99.92%)
'شكرا بزاف على المساعدة' → remerciements (99.45%)

📊 Model Performance

Metric	Score
Accuracy	92.8%
Precision	93.1%
Recall	92.8%
F1-Score	92.9%

Evaluated on 1,192 test samples

🛠️ Training Details

Base Model: UBC-NLP/MARBERTv2
Dataset: 5,960 labeled Darija sentences
Training: Fine-tuned for text classification
Features: Handles code-switching (Darija + French)

📄 Citation

@misc{mediani2025darija,
  author = {Mohammed Mediani},
  title = {MARBERT Fine-tuned for Darija Intent Classification},
  year = {2025},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/mediani/marbert-fine-tuned-darija-aicc}},
}

📜 License

Apache 2.0

Downloads last month: 16

Safetensors

Model size

0.2B params

Tensor type

F32

MohammedMediani
/

marbert-fine-tuned-darija-aicc