A practical guide to Singular Value Decomposition in Python
summary
Explains how SVD produces matrix factorization and item similarity recommendations, implements a simple movie recommender on the MovieLens data using Python, Jupiter notebooks and the Surprise SVD package, trains an SVD with ~100 latent factors, inspects the Q_i matrix and raw-to-inner ID mapping, predicts ratings by dot product and ranks similar movies with cosine distance, shows concrete examples (Star Wars ↔ Return of the Jedi, Pulp Fiction neighbors) and notes training is cheap on small MovieLens subsets.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 3:59Collaborative Filtering
- 7:52What is SVD
- 9:45Latent features
- 17:57Overview
- 18:50MovieLens
- 19:55How to train an SPD
- 22:19Where are the latent features
- 23:09The hidden attribute
- 23:38Example
- 24:05Prediction
- 25:12Comparing items
- 26:33Similarity ranking
- 27:46Conclusion
- 28:19Questions
related talks
01h 38mBuilding Recommender Systems Using Python
37mA primer on recommendation systems.
01h 31mHands on - Build a Recommender system
32mLet's talk about Recommendation Systems!
33mPicking What to Watch Next: Build a Recommendation System
46mBeyond Bag of Words A Practitioner’s Guide to Advanced NLP
38mRecommender systems with Tensorflow
17mLarge-Scale Parallel Recommender System in Julia