So tell me, what is your musical taste?
summary
A simple web app called Stratify extracts eight Spotify audio features (danceability, energy, loudness, acousticness, valence, tempo, etc.) via the Spotify Web API, averages a user’s top 50 tracks into an eight‑dimensional taste vector, reduces it to two dimensions with PCA to visualize clusters, and ranks candidate songs by cosine similarity for recommendations. Comparison shows Discover Weekly clusters near personal taste while Global Top 50 does not; note there are ~50 million tracks (~308 years to listen to all), so content+collaborative methods are needed.
AI-written summary · about
chapters (10)
- 0:00Start
- 3:37Music taste modeling
- 4:57Spotify Web API
- 6:40Island in the Sun - Weezer
- 8:11Modeling my music taste
- 9:20Box plot of audio features
- 13:37Measuring similarity
- 15:54Collaborative filtering (CF)
- 17:37Cold start and Long tail problem
- 18:48Next step Let's combine content based and collaborative filtering CFX
related talks
50mBetter Spotify Playlists through Music Theory & Discrete Optimization
10mInteractive Music Genre Exploration with Visualization and Mood Control
24mHow Spotify built an audience framework that lasts
38mmusic2vec: A smart music recommendation and discovery
35mDiversity and Data: Cases in the Music Industry
59mClosing Keynote: Paul Lamere
40mBuilding A Folk Music Recommendation System with LLMs
25mCombining Collaborative & Content Filtering in a Recommendation System for a Web-based DAW