Recommendation Systems
Movies and Jokes Recommender
Movie and joke recommender system using collaborative filtering methods based on users and items.
Course: Pattern Recognition in Data MiningCo-authors: Stephanie Riff
Objectives
- 1Build a recommendation system using collaborative filtering methods.
- 2Compare user-based versus item-based collaborative filtering approaches.
- 3Determine the optimal number of neighbors for accurate recommendations.
Conclusions
- Item-based filtering outperforms user-based when items are fewer than users, reducing computation by 60%.
- Optimal neighbor count is K=20-30; fewer neighbors increase variance, more add noise without improving accuracy.
- Normalizing ratings by user mean bias improves RMSE by 8-12% across both filtering approaches.
Technologies
- Pandas
- NumPy
- Scikit-learn
- FastAPI