RSS Swipr

A swipe-based RSS reader that learns your preferences. Swipe through articles to train a personalized ML model that recommends content you'll enjoy.

Quick Start

# Clone and setup
git clone https://github.com/philippdubach/rss-swipr.git
cd rss-swipr
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

# Run
python app.py

Open http://127.0.0.1:5000 in your browser.

Usage

1. Add RSS Feeds

1. Click Settings (gear icon)
2. Go to Feeds tab
3. Paste CSV or upload a file:

   name,url
   Hacker News,https://news.ycombinator.com/rss
   TechCrunch,https://techcrunch.com/feed/

4. Click Refresh Feeds to fetch articles

2. Swipe Articles

• Swipe right = Like
• Swipe up = Neutral
• Swipe left = Dislike

The app tracks your votes, link clicks, and reading time to build training data.

3. Train Your Model

Once you have enough votes (50+ recommended):

1. Export: Settings → Export → Download Training Data (CSV)
2. Train: Open the Google Colab notebook, upload your CSV, run all cells
3. Upload: Settings → Models → Upload the generated .pkl file
4. Activate: Click "Activate" to use your model for recommendations

4. Backup & Restore

To use your training data on a fresh install:

1. Keep your training_data.csv from the export step
2. On new install: import feeds, refresh to fetch articles
3. Settings → Export → Import Training Data (upload your CSV)

This restores your voting history by matching articles via URL.

How It Works

• Thompson Sampling: 80% exploit (best predictions), 20% explore (diversity)
• Hybrid Features: Combines text analysis with behavioral signals
• No cloud dependency: All data stays local in SQLite databases

Project Structure

rss-swipr/
├── app.py              # Flask server
├── src/                # Python modules (feeds, tracking, models)
├── ml/                 # ML pipeline and trained models
├── static/             # Frontend (JS, CSS)
├── templates/          # HTML
└── notebooks/          # Model training notebook

Requirements

• Python 3.8+
• Dependencies: Flask, pandas, scikit-learn, xgboost, feedparser

License

MIT