Hi,
How does similar artists work? Is it based on tags in Musicbrainz, or something else?
I’ve been adding lots of artists in an underrepresented on Musicbrainz genre (“dungeon synth”)
So, is there anything I could be doing to help generate similar artists for them?
Thanks…
5 Likes
So I guess nobody knows!
One of the “biggest” artists in this particular scene. Listenbrainz radio on easy mode - it says there are no similar artists and the playlist is all tracks by “Lord Lovidicus”
On medium & hard modes it gives 1 similar artist (but a different one in each case) and adds a couple of tracks by that artist
If more people click the plus sign next to genre tags, will it make more similar artists I wonder?
In that case I might try to start a bit of a project for this genre!
Hi!
Sorry, I didn’t see this post before. First, you can explore the similar artists data here:
Note that there is only one algorithm dataset live right now, so always use that same cryptic value for algorithm.
As for how the algorithm works:
- The listen stream for the past X days (9000 here) is broken into user’s streams and then each user’s listens for a window of 300 minutes is broken into a “session”. The basic idea is that tracks listened to in the same session are “similar”.
- For each possible pair of tracks, and each artist on both tracks (there may be more than one) each pair of artists its “score” is increased by 1.
- In the end all possible pairs are collected and pairs with a minimum score are kept. If a score is too high (e.g. the same track listened to over and over again) then that is capped at 50 counts contributed by this session.
So, when you see a score of “5,769” between two artists, it means that 5,769 track pairs, which translate into artist pairs, were found in the global listen stream.
That said, if the scores are too low, it means that not enough people are listening to that artist. The solution is to find other/new users who love those artists and get them to start listening to them too. Overall growth of our community will improve this data.
12 Likes
Oh wow, thank you! And thanks for the explanation, very clear. Will continue to evangelize about Listenbrainz in my scene!
4 Likes
