Recommender Systems

Slides in Keynote and PDF.

Content

• Neighborhood Methods

  • Motivation

  • Nearest Neighbors

  • Nadaraya Watson Style interpolation

  • Bias

  • Similarity between ratings

• Matrix Factorization

  • Latent variable methods

  • Stochastic Gradient Descent

  • Factors with bias

  • Aldous Hoover Theorem (exchangeability over matrices)

• Session Modeling

  • Users in context

  • User intent

  • Sequential content consumption (implicit user modeling)

  • Independent clicks, logistic model, sequential clicks, skip-click-view, with context

• Feature Representation

  • Latent factors from context

  • Cold start problem

  • Multiple sources

  • Homophily

• Hashing for recommendation

  • Matrix hashing

  • Approximation guarantees

  • Optimization

Supplementary material

• Collaborative Filtering with temporal dynamics


• Neighborhood factorization (note the insanely awesome URL of the paper …)

• Matrix Factorization for recommender systems


• CoFi Rank (collaborative filtering & ranking)


• Yehuda Koren papers


Video