Music Personalization At Spotify
14 likes•7,386 views
Spotify provides personalized music recommendations to over 100 million active users based on their listening history and the listening history of similar users. It utilizes various recommendation approaches, including collaborative filtering using latent factor models to create lower-dimensional representations of users and songs. Spotify also uses natural language processing models on playlist data and deep learning on audio features to power recommendations. Personalizing music at Spotify's massive scale across 30 million tracks presents challenges around cold starts, repeated consumption, and measuring recommendation quality.
![[1] http://benanne.github.io/2014/08/05/spotify-cnns.html
Deep Learning on Audio](https://image.slidesharecdn.com/musicpersonalizationatspotifyrecsys2016final-161101212337/85/Music-Personalization-At-Spotify-17-320.jpg)
Music Personalization At Spotify
- 2. Spotify’s Big Data ‣ Started in 2006, now available in 58 countries ‣ 100+ million active users, 35+ million paid subscribers ‣ 30+ million songs in our catalog, ~20K added every day ‣ 2+ billion playlists ‣ 1 TB of log data every day ‣ Hadoop cluster with ~2500 nodes
- 6. Personalization @ Spotify Features: Discover Discover Weekly Fresh Finds Home Radio Release Radar 5
- 7. Approaches ‣Manual Curation by Experts ‣Metadata (e.g: Label Provided Data, News, Blogs) ‣Audio Signals ‣Collaborative Filtering ‣ Hybrid
- 8. Latent Factor Models “Compact” representation for each user and items(songs): f-dimensional vectors
- 9. Latent Factor Models “Compact” representation for each user and items(songs): f-dimensional vectors Vidhya Rise .. . . . . .. . . . . .. . . . . .. . . . . .. . . . . .. . .. . .. . .. . . . ... ... ... ... .. mUsers Songs
- 10. Latent Factor Models “Compact” representation for each user and items(songs): f-dimensional vectors Vidhya Rise .. . . . . .. . . . . .. . . . . .. . . . . .. . . . . .. . .. . .. . .. . . . ... ... ... ... .. mUsers Songs User Vector Matrix: X: (m x f)
- 11. Latent Factor Models “Compact” representation for each user and items(songs): f-dimensional vectors Vidhya Rise .. . . . . .. . . . . .. . . . . .. . . . . .. . . . . .. . .. . .. . .. . . . ... ... ... ... .. mUsers Songs User Vector Matrix: X: (m x f) Song Vector Matrix: Y: (n x f)
- 12. Latent Factor Models “Compact” representation for each user and items(songs): f-dimensional vectors (here, f = 2) Vidhya Rise .. . . . . .. . . . . .. . . . . .. . . . . .. . . . . .. . .. . .. . .. . . . ... ... ... ... .. mUsers Songs User Vector Matrix: X: (m x f) Song Vector Matrix: Y: (n x f)
- 13. NLP Models on News and Blogs
- 14. NLP Models work great on Playlists!
- 15. Document : Playlist NLP Models work great on Playlists!
- 16. Document : Playlist Word : Song NLP Models work great on Playlists!
- 19. Music in Latent Space
- 20. Vectors “COMPACT” representation for users and items musical fingerprint. Normalized Song Vectors
- 21. Vectors “COMPACT” representation for users and items musical fingerprint. Normalized Song Vectors User Vector
- 22. Why Vectors? Encodes higher order dependencies Users and Items in the same latent space User - Item recommendations Item - Item similarities Easy to scale up Complexity is linear in order of latent factors
- 25. Ranking Similarity score can be used for ranking
- 26. Ranking Similarity score can be used for ranking Balance relevance, diversity, popularity, freshness
- 27. Ranking Similarity score can be used for ranking Balance relevance, diversity, popularity, freshness Heuristic based
- 28. Ranking Similarity score can be used for ranking Balance relevance, diversity, popularity, freshness Heuristic based MAB Interactions Impressions Clicks Streams
- 29. Music Personalization Data Flow
- 30. 18
- 31. Challenges Unique to Spotify Scale of catalog Music is “niche” Music consumption has heavy correlation to users’ context Repeated consumption of music is NOT so uncommon.
- 32. Challenge Accepted! Cold start problem for both users and new music/upcoming artists: Content Based Signals Real Time Recommendations Measuring Quality: Implicit: A/B Test Metrics Explicit: Feedback from social forums Scam Attacks: Rule based model to detect scammers Humans choices are not always predictable: Faith in humanity
- 33. What Next? ‣Personalization! ‣Content signals such as lyrics, audio, images ‣Expanded Catalog: Shows, Podcasts ‣New Markets 21
- 34. We are hiring!
- 35. Thank You! You can reach me @ Email: [email protected] Twitter: @vid052 23