KHALIDMRJ/Darija-Sentiment-Analysis

🇲🇦 Darija Sentiment Analysis تحليل المشاعر بالدارجة المغربية First open-source, end-to-end NLP pipeline for Moroccan Arabic (Darija) From raw web scraping → preprocessing → model training → live inference > *"40 million people speak Darija. Until now, AI couldn't understand a single word of their sentiment."* --- ## 📌 Table of Contents - The Problem - What We Built - Live Demo - Dataset - Data Analysis - System Architecture - Preprocessing Pipeline - Models & Results - Project Structure - Quick Start - Research Findings - Roadmap - Academic Context --- ## 🚨 The Problem No One Solved **Moroccan Arabic (Darija)** is spoken by over **40 million people** daily. It is a living, evolving language — a unique fusion of: - 🔤 **Classical Arabic** — grammatical backbone - 🇫🇷 **French** — embedded in everyday speech - 🏔️ **Tamazight (Berber)** — indigenous vocabulary - 🇪🇸 **Spanish** — in northern regions Despite this scale, **Darija is one of the most NLP-neglected dialects in the world.** | Challenge | Impact | |---|---| | No standard spelling | Same word = 5+ forms | | Mixed scripts | Arabic + Latin + digits | | No labeled datasets | Training impossible | | No preprocessing tools | Zero libraries exist | | Dialectal variation | Region-to-region differences | **The consequence:** Moroccan companies, researchers, and public institutions cannot perform automated sentiment analysis on the content their users generate every day — social media, reviews, news comments, customer feedback. **This project is the first complete answer to that problem.** --- ## 🎯 What We Built An **end-to-end NLP pipeline** covering every stage from raw data to live inference: ``` ┌─────────────────────────────────────────────────────────────────────────┐ │ FULL PROJECT PIPELINE │ │ │ │ 🌐 Web Scraping → 🧹 Preprocessing → 🏷️ Labelin …