CineWrap 2026 — Hybrid Movie Recommender
Hybrid movie recommendation system inspired by Netflix Wrapped that combines three signals: Collaborative Filtering based on SVD, Content-Based Filtering based on genre cosine similarity, and Popularity signal as a quality stabilizer.
Detailed Insights
Feature Engineering
Recency Weighting with exponential decay is applied to prioritize recent rating preferences. User profile vectors are constructed in 19-dimensions based on L1-normalized genres.
Collaborative & Content Based Filtering
Collaborative Filtering uses TruncatedSVD for matrix factorization (k=50) capturing 80.6% variance. Content-Based uses genre cosine similarity between user profiles and movie feature matrices.
Hybrid System
Combines scores from CF (0.50), Content (0.30), and Popularity (0.20) to mitigate cold start/sparse data, maintaining highly personalized recommendation relevance with only 4.0% overlap rate in testing.
Tech Stack
Key Results
- Hybrid: 0.5 CF + 0.3 Content + 0.2 Pop
- TruncatedSVD explain 80.6% variance
- Overlap rate 4.0% between users