thunderbolt250/rwanda-sentiment-analysis

# Rwanda Political Sentiment Analysis An end-to-end NLP pipeline that collects, preprocesses, labels, and classifies sentiment in tweets about Rwandan politics — with an interactive Streamlit dashboard. --- ## 📌 Table of Contents - Overview - Demo - Project Structure - Pipeline - Results - Key Findings - Tech Stack - Setup & Usage - Lessons Learned - Author --- ## Overview This project analyzes public sentiment on Twitter/X about Rwandan political topics using machine learning and NLP. It covers the full ML pipeline — from data collection to a deployed interactive dashboard. **Problem:** What is the overall sentiment of English-language tweets about Rwandan politics, and which words most influence that sentiment? **Approach:** 1. Collect tweets using keyword search (no expensive API — auth token approach via Scweet) 2. Clean and preprocess text 3. Label sentiment using VADER auto-labeling + 200 manual annotations 4. Train and compare two models: Logistic Regression baseline vs AfroXLMR transformer 5. Explain predictions using SHAP values 6. Deploy an interactive dashboard on Streamlit Cloud --- ## Demo > 🚀 **Live Dashboard → | Overview | Trends | Explainability | Live Predictor | |---|---|---|---| | Sentiment distribution charts | Monthly trend lines | SHAP word importance | Real-time tweet classification | --- ## Project Structure ``` rwanda_sentiment/ ├── data/ │ ├── raw/ # collected tweets (CSV) │ └── processed/ # cleaned, labeled datasets ├── src/ │ ├── collect_tweets.py # Phase 1: Twitter/X scraping │ ├── preprocess.py # Phase 2: text cleaning │ ├── label.py # Phase 3: VADER + manual labeling │ ├── train.py # Phase 4: model training │ └── explain.py # Phase 5: SHAP explainability ├── dashboard/ │ └── app.py # Phase 6: Streamlit dashboard ├── models/ │ ├── logistic_regression.pk …