yahia-frf/algerian_arabic_nlp_project

# 🇩🇿 Algerian Arabic NLP: Fine-Tuning BERT for Fake News Detection & Sentiment Analysis **A comprehensive NLP project fine-tuning state-of-the-art Arabic BERT models (DziriBERT & AraBERT) on Algerian Arabic dialect for binary classification tasks.** Key Features • Installation • Usage • Architecture • Results • Citation --- ## 📖 Table of Contents - Overview - Key Features - Project Structure - Installation - Dataset Preparation - Usage - Model Architecture - Training Pipeline - Evaluation Metrics - Results - Inference - Technologies - Future Work - Contributing - License - Citation - Acknowledgments - Contact --- ## 🧠 Overview This project addresses the critical need for **Natural Language Processing (NLP)** tools tailored to **Algerian Arabic (Darja)**, a unique dialect that blends Modern Standard Arabic, Berber, and French influences. We tackle two fundamental NLP tasks: ### 🎯 Tasks 1. **📰 Fake News Detection** Binary classification to identify misinformation in Arabic news articles and social media content. 2. **💬 Sentiment Analysis** Binary sentiment classification to understand positive/negative emotions in Algerian Arabic text. ### 🔬 Research Motivation Algerian Arabic presents unique challenges for NLP: - Limited labeled datasets - Code-switching between Arabic, French, and Berber - Dialectal variations distinct from Modern Standard Arabic - Underrepresentation in existing Arabic NLP models This project demonstrates that **dialect-specific models (DziriBERT) can outperform general-purpose Arabic models (AraBERT)** on Algerian content, validating the importance of regional language modeling. --- ## ✨ Key Features - 🤖 **Dual Model Comparison**: DziriBERT vs AraBERT v2 - 🎯 **Two Classification Tasks**: Fake news detection & sentiment analysis - 📊 **Comprehensive Evaluation**: Accuracy, F1-score, Precision, Recall - 🚀 **Production-Ready Training**: Early stopping, learning rate scheduling, gradient accumulation - 💾 **Google Drive Integra …