This article includes the following methods: Collaborative Filtering, Matrix Factorization, Logistic Regression, Gradient Boosting Decision Tree + Logistic Regression, Factorization Machines, Field-aware Factorization Machine, Deep Crossing, Wide & Deep, Neural Graph Collaborative Filtering, and DRN.
1. Collaborative Filtering (CF)
- In 1992, CF was first developed by Xerox Oalo Alto Research Center for their employees to handle a huge stream of incoming emails. CF enables users to subscribe only to those lists of interest to them.
- In 2003, Amazon published the paper “Amazon.com Recommenders Item-to-Item Collaborative Filtering”.
1.1 User-based CF
Example:
- Say an e-commerce website (e.g., Amazon) has 4 products: a certain brand of game controller, magazine, book, and TV.
- User X visits Amazon. Amazon’s RecSys need to determine if it should recommend TV to X. In other words, RecSys needs to predict if X likes this brand’s TV. To do so, it looks at the historical rating that all users gave to other products they bought.
- Green thumb-up means a good rating, red thumb-down means a bad rating. Then we have a matrix with users, products, and ratings. For the…