Anime Recommendation System
Scalable recommendation engine using collaborative filtering and SVD on a large MyAnimeList dataset.
Project Snapshot
Built an anime recommendation system comparing popularity-based models, user-based collaborative filtering, and SVD-based matrix factorization on a large, highly sparse MyAnimeList dataset. Demonstrated that SVD latent-factor modeling achieves the best predictive accuracy and scalability.
Problem
With thousands of anime titles available, users face difficulty discovering content matching their preferences. The challenge was building a recommendation engine that handles extreme data sparsity while producing accurate, personalized suggestions.
Approach
- Implemented popularity-based baseline models
- Built user-based collaborative filtering using similarity metrics
- Applied SVD-based matrix factorization to learn latent user-item factors
- Evaluated all models using RMSE on held-out test data
Tech Stack
Python, Pandas, NumPy, Scikit-learn, Surprise, Jupyter Notebook
Results
- SVD achieved the best predictive accuracy with RMSE ≈ 1.16
- Outperformed traditional collaborative filtering in both accuracy and scalability
- Demonstrated effective handling of large sparse rating matrices