Spotify Playlist Recommendation using Matrix Completion with Alternating Minimization. implementation of the theoretical work of (Hardt, 2014) https://arxiv.org/abs/1312.0925.
This project seeks to implement matrix completion for an artist recommendation algorithm for specific users? Playlist download tool for Spotify: Exportify https://exportify.net/
- How do recommendation algorithms work, e.g. Spotify and Netflix?
- What new artists should be recommended for users, given we have a general sense of their music taste?
- Which users have the most similar music taste, considering what artists were recommended to them?
Normalized values by row to find recommendations for each user ensure that each recommendation value was on a consistent scale.
- Top artist for each user has been scaled such that the max value is 1.0
- Artist that has the value of 1.0 is the artist that shows up the most in their playlists
Notably, we used a small dataset of only:
- 2808 artists
- 125 playlists
Which resulted in the optimal parameters of k = 58 and T = ~200 for playlist dataset, and k = 10 and T = ~10 for the user dataset.
PCA of user preferences by Euclidean Distance
Genre Analysis Post-hoc:
