tarekAeb/dziribert-algerie-telecom-v1

DziriBERT Algérie Télécom Classifier

This model is a fine-tuned version of alger-ia/dziribert for classifying Algerian telecom customer comments into urgency levels.

Model Description

• Base Model: DziriBERT (BERT for Algerian Arabic)
• Task: Text Classification
• Languages: Arabic, French, Algerian Darija
• Classes: 5 urgency levels
  • High_Urgency: عاجل جداً
  • Medium_Urgency: متوسط الأهمية
  • Low_Medium_Urgency: منخفض-متوسط
  • Low_Urgency: منخفض الأهمية
  • No_Urgency: لا توجد أهمية

Usage

from transformers import AutoTokenizer, AutoModelForSequenceClassification
from transformers import pipeline

tokenizer = AutoTokenizer.from_pretrained("tarekAeb/dziribert-algerie-telecom-v1")
model = AutoModelForSequenceClassification.from_pretrained("tarekAeb/dziribert-algerie-telecom-v1")

classifier = pipeline("text-classification", 
                     model=model, 
                     tokenizer=tokenizer)

# Example
result = classifier("الانترنت ما يخدم من البارح")
print(result)
# [{'label': 'High_Urgency', 'score': 0.89}]

Training Data

• Dataset: 9,771 customer comments from Algerian telecom social media
• Sources: Facebook, Instagram, Twitter posts
• Preprocessing: Text cleaning, normalization, augmentation

Performance

• Accuracy: 87%
• F1-Score: 85%
• Training Strategy: Layer freezing, hyperparameter optimization

Demo

Try the live demo: Comment Genie