I have added songs that I am currently listening to into a monthly playlist since Summer 2015. I have also kept a record of every song that Spotify has recommended to me via their Discover Weekly service since mid-2016. Using these datasets I have performed a rough analysis of Spotify's prediction model for my music taste. All the code that was used to achieve this is located in this repository (https://github.com/harritaylor/discover-weekly-analysis).
Spotify provides an API available to developers in order to make applications using the platform. I was able to query this API using the Requests library in python for song features of all the songs in both the discover weekly archive playlists and my saved favourites playlists. I then combined the datasets and introduced another variable, "saved" in order to distinguish which songs I had personally added to playlists, and which songs Spotify had recommended to me. I also removed all songs that appear in both playlists from the Discover Weekly Archive playlist, as these are songs that Spotify successfully recommended to me that I like. I then saved the data to a CSV file entitled "spotify_dataset_spring18.csv".
Using R's hist feature, I gained a general overview of my preferences in songs, and Spotify's idea of song features that I like. There are some features that Spotify gets spot on, and some others that are way off.
