Help with Calculating Spotify Profile Matches for a Scientific Experiment

[Eliahgrgi]

Hi everyone,

I’m currently working on my Bachelor’s thesis and I want to calculate the match between Spotify profiles to study its influence on relationship satisfaction. The idea is to have two people authenticate via the Spotify API, and then I analyze their listening data (Top Songs, Artists, Genres, etc.) to create a "match score."

My questions are:

1. Metrics: What metrics are best for calculating similarity between two users? I’ve been thinking about using Jaccard Index (for genres or artists) and Cosine Similarity (for audio features). Has anyone worked on a similar project?

2. Automation: Is there a way to replicate the Spotify Blend logic or use similar functions via the API? I would like to automate this match calculation.

3. Playlist Creation: How can I automatically create a playlist with the best matching songs from both users? I’m currently using Python and the Spotipy library.

4. Scaling: My goal is to provide this feature to multiple participants in an online experiment. Are there any best practices for integrating Spotify data into web apps (e.g., with Flask or Django)?

I’d appreciate any tips or resources that could help me implement this. Also, if anyone knows how I could contact Spotify directly to learn more about their algorithms (e.g., behind the Blend feature), that would be really helpful.

Thanks in advance for your support!

[ReubenF]

Unrelated to your question, but just FYI access to the audio features endpoint was removed for new/existing (development mode) apps: https://developer.spotify.com/blog/2024-11-27-changes-to-the-web-api