1camelea/jumia-sentiment-analysis

Sentiment analysis of Jumia Morocco reviews — French & Darija NLP # 🛒 Jumia Morocco — Sentiment Analysis > **NLP project** | TF-IDF + Logistic Regression | French & Darija (Moroccan Arabic) --- ## 📌 Overview This project builds a **sentiment analysis classifier** for customer reviews scraped from **Jumia Morocco** (jumia.ma) — the leading e-commerce platform in Africa. Reviews are written in a realistic mix of **French**, **Darija** (Moroccan Arabic dialect in Latin script), and **Arabizi**, making this a unique and challenging NLP task that reflects real-world Moroccan digital content. **Sentiment labels** are derived automatically from star ratings: | Stars | Label | |-------|----------| | ⭐⭐ | Negative | | ⭐⭐⭐ | Neutral | | ⭐⭐⭐⭐⭐ | Positive | --- ## 🗂️ Project Structure ``` jumia-sentiment/ ├── data/ │ └── raw_reviews.csv # Scraped / sample reviews ├── src/ │ ├── scraper.py # Jumia Morocco web scraper │ ├── generate_sample_data.py # Generates realistic sample data │ └── train_model.py # Full NLP training pipeline ├── outputs/ │ ├── sentiment_model.joblib # Saved trained model │ └── figures/ │ ├── eda_overview.png │ ├── confusion_matrix.png │ └── top_features.png ├── requirements.txt └── README.md ``` --- ## ⚙️ Installation ```bash git clone github.com cd jumia-sentiment-analysis pip install -r requirements.txt ``` --- ## 🚀 Usage ### Step 1 — Get data **Option A: Scrape live data from Jumia Morocco** ```bash python src/scraper.py ``` **Option B: Generate a sample dataset (for testing)** ```bash python src/generate_sample_data.py ``` ### Step 2 — Train & evaluate ```bash python src/train_model.py ``` --- ## 🧠 NLP Pipeline ``` Raw Review Text │ ▼ Preprocessing ───────────── • Lowercase • Remove URLs, punctuation • Remove French + Darija stopwords • Preserve Arabizi digits (3 = ع, 7 = ح, 9 = ق) │ ▼ TF-IDF Vectorizer ───────────────── • Unigrams + Bigrams • max_features = 10 000 • sublinear_ …