- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page
How I make the Spotify algorithm work for me
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I used to be more of an album person. But over the years, I've found that having a playlist with hours of music that I can just put on shuffle is really enjoyable, and I've sort of cobbled together a strategy that I think works pretty well to get Spotify to surface new tracks that I'll like. If you're a shuffler like me, you might find some value in this. Full disclosure, I've never worked for Spotify, and I don't actually know how the Spotify algorithm works, but over the past five years I've been able to find hundreds of hours of new artists and tracks that I love, and I think it's at least in part because of this method. If you're interested, here's how I suggest you can do it too:
Step 1: Make a playlist that has as much music in it that you know you already like. You can put full albums in here, but I recommend that you stick to individual tracks if you want to get the best results possible (and by best, I mean the closest to mine, which are obviously the best). Ideally, you want several hours worth of individual tracks that you really like. If you don't like that much music already, I don't know what to tell you. Talk to your friends. Get some recommendations. Listen to the Beatles. Whatever. Otherwise, if you've got this playlist ready to go, name your playlist something clever or creative, like Baby's First Algorithm Trainer, and move on to step 2.
Step 2: Over the next two months, try to listen exclusively to your Trainer playlist on shuffle. If you just have to listen to other music for some reason (like it's the holidays or some other lame excuse), then try to do it on a different service. Remember, you're training an algorithm. Any deviation will be noise in the signal. For example, I typically like to listen to movie scores and other instrumental work when I am writing fiction. However, that's not the kind of music I like to listen to when I'm doing literally anything else, so I actually created a separate Spotify profile on which I listen to that kind of stuff on. Another example is if you have kids and let them use your Spotify profile. If you do this, you may never get the algorithm to work right for you. Just too much noise (looking at you Minecraft covers).
Step 3: After a month or two of listening to your Trainer, you'll want to start listening to your Discover Weekly. I suggest waiting this long just to ensure that you've given ample time to properly training your algorithm. By the second or third month, if you've been listening to your Trainer regularly, Discover Weekly should have some good stuff. This is where the magic really starts to happen. But before you can really start listening to your Discover Weekly, some infrastructure first. Create a new playlist. We're going to call it "Recent Discoveries." I mean you can call it whatever you want, but that's what I'm going to refer to them as. Once you've got that made, we're ready to start what I call, "The Workflow."
Step 4: You're going to want to get in a weekly rhythm here. On Mondays, when Discover Weekly gets populated, you're going to want to listen to the whole playlist straight through in order. Every time you hear a song you like (in my case I know this is the case because I'm involuntarily tapping my foot), you throw that sucker into your Recent Discoveries playlist (RD from this point on). After you finish with the first play, then you play the Discover Weekly again, but this time on shuffle. It just helps me make sure I haven't missed anything. Additionally, if you hear a song that you enjoy via non-Spotify means, like on the sound system at a coffee shop, or recommended by a friend, throw it into the RD.
Step 5: Once your RD playlist has a couple hours worth of music in it, you'll want to start adding it in to your regular rotation. So now, you're listening to your Trainer on shuffle sometimes, and your RD on shuffle sometimes. This will keep the algorithm tethered to your core preferences (via Trainer), while expanding it a bit (via RD).
Step 6: As you are listening to your RD, you might find yourself putting a particular track on repeat. That's good. If you're a binger like me, you can use that to help train the algorithm even more. Any song you catch yourself listening to more than once in a row, put into another new playlist we're going to call Discoveries on Repeat (DoR going forward). I know Spotify automatically creates a playlist of "On Repeat" songs, but I prefer to do it manually because I like to control what's in there at any given time. Also I've been doing this longer than Spotify has had an On Repeat playlist so I feel like I get to take credit for that completely original idea.
Now we've got three playlists to listen to regularly in order to train the algorithm. The original Trainer, the RD, and the DoR. In terms of the amount, it's a balance at first, and you'll want to feel out your own rhythm. For me, I really like songs that I can listen to over and over and obsess over, so I try to train Spotify to find me those by listening to my DoR playlist as often as possible. Over time, you'll need to listen to your Trainer less and less, until eventually it will morph into a new playlist that I'll discuss... uhh... right now.
Curation: In order to keep your algorithm results fresh, you're going to want curate both your RD and DoR playlists. For me, I try not to let RD get longer than 8-10 hours of music, and DoR under 3 hours. Why those? It's based on how I personally listen to music. During the work day, I like to listen to RD (hence 8-10 hours). When I'm driving, I like to listen to DoR (hence under three hours). That said, I will often listen to DoR while I'm working, and RD while I'm driving. The point though, is to keep the list of songs you have recently discovered curated, and your repeat songs even more so.
Once I get over the threshold, how I cull is different for RD vs DoR. For DR, I will listen to the tracks that have been in the playlist the longest, and then move them into the Trainer playlist, removing them from RD. See now, if you get to a point where your RD and DoR feel stale, you can always go back to your Trainer which will have an ever-growing number of tracks you've liked at some point. It's like a palate cleanser. How I cull DoR is just based on whenever I get tired of listening to a track on repeat. Once that happens, I throw it in a different playlist I call "Repeat Archives." This is another playlist you can get mileage out of when your others feel stale.
Once you're all set up, you're going to have at least three playlists in the workflow: the Trainer (all the stuff you like, basically, unlimited number of tracks), the Recent Discoveries (8-10 hours worth of stuff you like that you've heard recently), and Discover Weekly (what Spotify thinks you will like). If you do the repeat thing, you'll also have the Discoveries on Repeat playlist (under three hours) and Repeat Archives (everything you've had on repeat).
I have recently actually taken this one step further, because I realized that the tracks I listen to on repeat have stratified one level deeper. Some songs I listen to on repeat a few times, and others I will listen to for literally hours. I call these obsessions, and made an archive playlist accordingly. So if we're thinking of this as quality, my ranking goes like this: Obsessions > Repeat Archives > Trainer. And so I generally find myself listening to it in roughly corresponding amounts as well. And there's an inverse relationship with how many tracks are in each. Obsessions has the fewest, Repeat Archives more, and the Trainer the most.
Here's an example of how this might shake out in any given week:
1) Listen to Discover Weekly for 2-4 hours (twice through), pulling out tracks I like into RD. I don't know how or whether listening to the Discover Weekly affects what Spotify puts into this playlist, but in case it does, I try to limit (because often there are plenty of tracks in there I don't actively like, and some I actively dislike).
2) If RD is longer than 10 hours, listen to the oldest stuff in it, then move it to the Trainer until RD is under 10 hours. Then shuffle RD for a good bit. If a track gets intentionally put on repeat, put it in DoR.
3) If DoR is longer than three hours, identify the tracks that have been in it the longest or feel the stalest. Move them into the Repeat archives. If it was an obsession song, move it into the Obsessions playlist as well.
4) Over any given week, listen to each category (like/repeat/obsession) roughly equally. If ten hours is spent on each category, because the number of tracks in each category corresponds to quality inversely, effectively you are prioritizing stuff you like the most.
From there, you can do other interesting things. I like to create manually curated mood playlists that I sometimes get very into. If I get obsessed with a track Spotify has gifted me, I will sometimes go look up the artist and see if they got more stuff I like. But usually what I'll do there is make a playlist that has a bunch of songs from a bunch of artists I liked one track from, and I'll shuffle that so that I don't get bogged down in one album (I like shuffle, okay?).
Bottom line, you want to be listening to as much different stuff that you like as much as possible for Spotify to identify more stuff you might like. Hope some people find this useful! If not, I probably won't ever find out! Happy listening!
The End.
- Labels:
-
App & Features
