AI-Powered Gift Recommendations
Works in any language · Dutch, German, French, Chinese, English
Your saved gift ideas with notes
Click the Save button on any gift to add it here
Selection: Amazon Reviews 2023 dataset filtered to 9 gift-relevant categories including Electronics, Toys, Beauty, Home & Kitchen, and more.
Preprocessing: Mapped 3.1M users and 135K products to numeric IDs, normalized prices to comparable scales, removed duplicates and handled missing values.
Train/Test Split: 80% of ratings (2.4M) used for training, 20% (600K) held out for testing model accuracy on unseen data.
Cross-validation: K-fold validation tested different hyperparameter combinations to find optimal settings without overfitting.
Model: Matrix Factorization with SGD optimization.
Parameters: 20 latent factors capture user/product characteristics, learning rate 0.005 ensures stable convergence, regularization 0.02 prevents overfitting.
| RMSE (Test) | 0.87 |
| MAE (Test) | 0.68 |
| Training Epochs | 15 |
| Convergence | Achieved |
RMSE 0.87 means predictions are on average 0.87 stars off from actual ratings (on a 1-5 scale). MAE 0.68 is the average absolute error. Both are strong results for recommendation systems.
Optimization Finished: RMSE stopped improving significantly after epoch 12-13.
Diminishing returns: Going from epoch 14 to 15 only improved RMSE by 0.01.
This project includes a fully functional Amazon Associates affiliate integration. Every product link contains our unique affiliate tag, enabling commission earnings on qualifying purchases.