I'm a big fan of this for how simple it is. How anyone that has used Spotify for years, knows how bad their random song algorithm is. Can you elaborate more on your 'simple custom algorithm'? I 'recycled' a large playlist with 650+ songs and I'm listening to songs I almost never listen to, I love it.
Thanks! Glad you've found it useful :). The algo is quite simple but a little bit hard to explain so I'll do my best...
The aim is to find similar artists to the ones you have selected based on genre (every artist on Spotify is associated with multiple genres).
1. The tool pre-calculates the number of artists from your playlists that are associated with each genre (e.g. "jazztronica: 3" indicates that there are 3 artists in your playlists in the genre jazztronica).
2. We compute a threshold called artist_thresh which is a function of the total number of artists across all your imported playlists and the "Similarity Threshold" set in the tool.
3. For each artist you select in the tool, we find all of their associated genres which have an artist count BELOW our artist_thresh.
4. For each of the genres selected in the previous step (as well as any genres manually selected in the tool), add all the associated artists to our output playlist.
5. Finally apply the "Release Years" and "Include Explicit Tracks?" filters and output the playlist.
The underlying assumption here is that rarer genres are more specific and therefore better indicators of similarity. E.g. I am more confident that two artists that share the genre "uk alternative hip hop" are similar than if they only share the genre "rap".
