Updates

[mf-caglar]

October 10

• I did the initial cleaning on datasets. Now I will create a metadata table and reduce the current tables into last forms as much as possible. Then I will obtain Python equivalents of the codes and review the notebooks on data we have and maybe replicate them on combined data. There are around 500k artist-song pairs. Let's discuss things on Sunday. Especially audio processing stuff.

[mf-caglar]

I finally exported combined tables and further combined and reduced them in google colab.

• data is here
• I created a notebook where I discover spotify API a little. it is really easy to start to use it. we can do several experiments on song data including audio, yearly trends and so many another things. that's fun and great to use. notebook is here

here is my first idea:

I am proposing a genre-based exploration of music, where we:

• Start with an artist (e.g., The Beatles), which is associated with several genres (e.g., Rock, Pop, British Invasion).
• Branch out from each genre by finding other artists within that genre based on a metric (like popularity or influence).
• Move to new genres as we explore artists and songs, expanding the tree by finding connections through related genres or artists.
• Explore whether the genre tree is interconnected: Can we reach completely different genres (e.g., Classical, Hip-Hop) by continuously stepping through artists and genres, or do we hit isolated "genre islands"? So genres are trapped at some point?
• Oh right of course "really" exhausting it maybe is impossible. But under a huge popularity threshold?
• Based on this we can create a "trendy" genre families.