Spotify Automatic Playlist Continuation

TLDR: This project consisted in building and comparing three different recommender systems (user-based, item-based collaborative filter and a neural network approach) on a dataset of 100K playlists and more than 600K songs. The recommender systems are used in order to continuate a playlist with related songs. I used PySpark to handle the large amount of data, and Petastorm to create a Pytorch dataloder from that data in an efficient manner.

Presentation • GitHub Repo

Ops, details of the project are still work in progress!