Growth Accelerator — Recommender System

Implicit feedback recommender system using truncated SVD to suggest products to users based on purchase history.

Problem → Approach → Results → Next Steps

• Problem. E‑commerce platforms need to increase conversion and basket size by recommending relevant products without explicit ratings.
• Approach. Generated a synthetic transaction dataset and constructed a user–item matrix of implicit feedback. Trained a TruncatedSVD model to factorize the matrix; normalized user and item factors; and ranked items via dot‑product scores. Provided a CLI to return top‑N recommendations for a user.
• Results. Hit@5 on a held‑out test set is approximately 0.60–0.65; training completes in under 2 seconds on a laptop with a compact artifact (< 5 MB).
• Next steps. Replace SVD with implicit ALS (with confidence weights); incorporate item content features to handle cold‑start; add diversity/novelty re‑ranking; and schedule periodic retraining.

Dataset

The script data/generate_transactions.py simulates a set of users and items, producing an implicit feedback matrix (user_item.csv). You can control the number of users, items, and transaction density via command line flags.

Installation

git clone https://github.com/yourname/growth-accelerator-recsys.git
cd growth-accelerator-recsys
python3 -m venv env
source env/bin/activate
pip install -r requirements.txt

Usage

Generate Synthetic Transactions

python data/generate_transactions.py --users 1000 --items 100 --output data/transactions.csv

Train the Recommender

python src/train.py --data data/transactions.csv --model svd --factors 50 --output models/

Get Recommendations

python src/infer.py --model models/svd.joblib --user_id 42 --n 5

The inference script outputs the top N items and their scores.

Project Structure

growth-accelerator-recsys/
├── data/
│   └── generate_transactions.py
├── src/
│   ├── train.py
│   ├── infer.py
│   └── …
├── models/
├── tests/
├── requirements.txt
├── .gitignore
├── .github/workflows/python-ci.yml
├── LICENSE
└── README.md

Contributing

Issues and PRs are welcome. Please add tests for new features.

License

This project is licensed under the MIT License.