Using Spotify API for Audio Data and User Feedback

[Izusl]

I'm currently working on a project that involves using the Spotify API to create a music recommendation system based on user feedback and audio characteristics. Specifically, I want to explore the following approach:

1. Use the Spotify API to retrieve songs within a certain valence (or other audio data) threshold (for example, 0.2 to 0.4).
2. If a song with a valence of 0.35 is played and the user dislikes it, I would like to adjust the valence value downward to 0.25 and continue iterating until we find a valence level that the user prefers.
3. The goal is to effectively create a feedback loop that personalizes recommendations based on the user's musical tastes without violating any terms of service.

My questions are:

• Is this approach compliant with Spotify’s terms of service regarding API usage?
• Are there any specific limitations I should be aware of when it comes to using audio features for machine learning purposes?
• Would this method be considered as “ingesting” Spotify content into a machine learning model?

[Izusl]

Something like this maybe stored:

{
"song_title": "Sad Melody",
         "artist": "Melancholy Artist",
         "valence": 0.6,
         "user_feedback": "disliked", 
}

[tomjaimz]

I don't think you need to use machine learning for this, or an LLM.  It takes a huge amount of data to train an LLM, and I don't think you'd create anything near that volume of data from your use case.  Of course, training an LLM violates the policy too.  Just build an app that creates a user feedback loop and adjusts valency of music until you find the user's desired level.