Using Spotify's SPOTIPY API, I pulled attributes describing thousands of songs that me and 3 other friends had 'favorited' on Spotify, with the goal of comparing our musical tastes using Machine Learning. I used the OneClass Support Vector Machine (OneClassSVM) algorithm from Python's scikit-learn library to visualize each of our music 'taste profiles' based on several song attributes that Spotify provides including:
- Beats Per Minute (BPM) — The tempo of the song.
- Energy — The higher the value, the more energetic.
- Danceability — The higher the value, the more danceable the track
- Loudness — The higher the value, the louder the song (in dB).
- Valence (aka positivity) — The higher the value, the more positive the mood.
- Length — The duration of the song.
- Acousticness — The higher the value the more acoustic the song
- Instrumentalness — the higher the value, the fewer the vocals
- Liveness — higher value indicate the presence of a live audience
The code to extract the songs using the Spotipy API can be found in the Jupyter Notebook files that start with "extract_..."
The code to analyze the songs can be found in the Jupyter Notebook file titled "eda_top_tracks"
Below are some charts comparing 4 individual's Music Tastes. Can you spot which one belongs to a Gen Z? :)
These charts compare the attribute distributions for the 3 individuals' favorite Spotify songs
Finally, this is a visualization of an individual's music taste profile using Principal Component Aanalysis and OneClassSVM algorithms. The orange-shaded area represents the music taste boundary that the algorithm learned. I added labels to show my favorite Taylor Swift songs.
