Simple Movie Recommender Using SVD

Abstraction: SVD-based movie recommender using cosine similarity on principal components

Key points:

• Uses MovieLens 1M dataset; SVD decomposes normalized ratings matrix A = U S V^T
• Top k columns of V^T (principal components) represent the k most informative rating directions; k=50 used in example
• Cosine similarity over the reduced V^T space finds movies with similar rating patterns across users
• Toy Story recommendations include Toy Story 2, Babe, Bug's Life — genre overlap without explicit genre features
• SVD and PCA via covariance eigenvectors yield identical results: V^T columns equal eigenvectors of A^T A
• SVD preferred over manual PCA: singular values are pre-sorted; Facebook's fast randomized SVD is faster still

Connections: Movielens · Singular Value Decomposition · Collaborative Filtering · Recommender Systems · Dimensionality Reduction

Source: https://alyssaq.github.io/2015/20150426-simple-movie-recommender-using-svd/