Personalization Project

This project was created during the class Personalization Theory and Application, taught by Brett Vintch at Columbia University, Fall 2017

Collaborators

Collaborators on this project are:

  • Jan-Felix Schneider
  • Manksh Gupta
  • Andres Potapczynski
  • Mohamed Maskani Filali

Overview

In this final project we explored the use of a biased matrix factorization model to improve the prediction of accuracy ratings for a music recommender system. To train the system we used data from the music service Deezer. As a baseline we were using neighborhood based colloborative filtering.

How to read this repo

The final report can be found here.

It contains a detailed description of the data, our methodology and our results.

The different models (baseline, matrix factorization, neighborhood filtering) can be found here