Recent updates

Similar Artist Filtering

I've been hoping to add this feature to our platform for a few years now, and I'm excited to announce that it's finally available! In addition to filtering curators by genre, you can now also filter them based on the specific artists they enjoy.

Table of contents:

  1. Choosing your similar artists
  2. How it factors into the "Genre Match" score
  3. The data / algorithm
  4. Next steps

Feel free to watch the video below if you're not so into reading. Or... do both :)

Choosing your similar artists

I've done a bit of an overhaul around the submission flow to work in the new "Similar artists" section below. The new flow is:

  1. Select your song
  2. Choose targeting (curators/influencers) and credit preferences (premium/standard)
  3. Provide details about your similar artists (new), genres, and moods
  4. Indicate feedback preferences, copyright permissions, and communication details (new)
  5. Find curators and submit!

The main difference is that sections 3 & 4 above used to be on one page, but are now split out. This extra step (#4) allows me to run a whole bunch of complicated calculations for the curators feed in the background while you're busy setting your options.

  • Here's the similar artists prompt:
  • The "search" function of the similar artists page is powered by Spotify's API, and the results are (for the most part) pretty good
  • Once you've made it to the list of curators, you'll see your selections in the sidebar. You can edit them from there, if you wish

How it factors into the "Genre Match" score

For the last few years I've emphasized the "Genre Match" score as the most-important metric. Historically, this has included a curator's genre and mood history. Today we're moving from "2D to 3D" by adding a third component.

  • Genre history:
  • How often you've approved the genre relative to your other genres and relative to peers
  • In the case of Spotify listeners, we also look at how much their playlists "focus" on the genre(s) in question
  • This remains the core component of the "Genre match" score, and sets the baseline from which adjustments are made based on the next two factors
  • Similar artists:
  • The data is currently dependent on Spotify playlists, and not everyone actively maintains playlists (eg, our bloggers)
  • I've made sure that a lack of data doesn't mean you get *punished* - playlists and blogs are being treated differently here, and then brought together at the end
  • Matches are not just based on the artists chosen by a submitter, but also the "fans also like" for those artists - doing two levels deep. I explain this a bit in the next section
  • Moods
  • This one has a smaller impact than the other two, but if we see that a curator doesn't want a specific mood, we tone down their genre match score. Likewise, if it's one of their favorite moods, we give their genre match score a slight lift

The data / algorithm

There are two primary data sources to be used here. One of them is done, the other one is in progress.

Spotify playlists (done)

We look at all the tracks in that curator's playlists and try to find overlaps between the artists you've chosen.

  • For every exact match they have with your search, they earn +5 points
  • We then look at the top 20 "similar artists" for the artist you've searched. Each song featured by one of these artists earns the playlist +2
  • We then go one step deeper and analyze the next level of "secondary similar artists" from step #2, and for each song featured the playlist earns +1
  • You can get a visualization of this when you click the "Similar artists" not that shows up under a curator's profile

This methodology means that for each artist provided by a submitter we're actually considering up to 400 different artists (1 x 20 x 20), and it helps us avoid a problem that a couple folks have expressed concern about: "If I only feature indie artists, and someone searches by a massive artist, what happens?"

Once we have a "score" for each curator, we stack rank them, and give a boost to the Genre Match score (maximum of +2.5 out of 10). So for example, the person at the top of the list will get +2.5 points on their genre match score, while someone in the middle might get 1 point extra. This means someone's genre match score could hypothetically move from 5/10 to 7.5/10.

Approval history (in progress)

Not everyone has a playlist and we've got a treasure-trove of historical data on SubmitHub for the majority of our curators. I'm in the process of writing a script that associates the right artists with the right approvals so that they can essentially be treated as "tracks in a playlist". This will be factored separate to the playlists and will need to avoid double-counting (ie, someone with lots of approvals of an artist also likely has those songs in their playlists).

Next steps

Not a huge list. Here goes:

  • I need to associate artist Spotify IDs with SubmitHub profiles so that I can use the "Approval history" methodology above
  • Once that's running, I want to add a filter to the sidebar that lets you only view curators who match the artists you've selected
  • Use the artists you've selected as "similar" to try make suggestions about genres
  • This is a hard one! Gonna save it for later

I might add more to that. Who knows. In the meanwhile, I'm crossing this one off my 2023 Wishlist 😁