Hung Up: the magic of Spotify’s algorithm

A Spotify Discover Weekly playlist is delivered to every one of its 113 million users every single week.

And it’s like the damn thing knows us. The algorithm rounds up a mix of old and new music we haven’t heard before, mysteriously tapping into our subconscious and delivering undiscovered delight with a spine-chilling hit rate.

This thing is a threat. I’m not kidding. The other day my husband excitedly put on a track from his Discover Weekly that had knocked his socks off.

“I looked at the artist — and I saw it was later-era Madonna! WHAT! And it’s a great song!”

He proceeded to put on Hung Up, the lead single from Madonna’s Confessions on a Dancefloor.

Despite the song sitting at number one for three weeks in November 2005, selling nine million copies worldwide and being one of the biggest selling singles of all time, it had totally passed him by. This is because Sam paid as much attention to the charts then as he does now — claiming to find commercial pop unbearably shit even in the mid-2000s.

So how the hell did Spotify know to throw that poptastic curveball into his playlist, otherwise full of death metal, thrash, rap, electro and dream pop bands from Brooklyn with trendy haircuts? How did Spotify know Sam would love THAT particular mainstream Madonna song?

Spoiler — it turns out nobody really knows the answer. Or if they do, it’s so brain-knottingly complicated everybody would fall asleep halfway through reading it. And anyway, sharing their mysterious secret recipe would surely open Spotify up to competition from all the other platforms desperately trying to achieve personalisation on a deeply intimate level that not even a wife could accomplish.

But we do know some things.

1. It steals from other people’s playlists

In the nicest possible way, of course. Spotify combs the billions of playlists created by other users on the platform. When it spots a few songs on somebody else’s playlist that are the same as yours, but they’re listening to another artist or song you haven’t heard? Bingo. It might end up in your Discover Weekly list.

2. It profiles your listening taste

— into seriously distinct categories. Neo-soul, synth-pop, power-pop, push-pop (probably), it gathers them all up and uses them to build an individual taste profile. It then looks for users with profiles that align to yours to see what they’re listening to that you’re not yet.

3. It picks up on what’s trending

If other users are out there heavily playlisting something that aligns with your taste, you could well find it appearing on your Discover Weekly too. If you find your list has crossover with your friends, this could be the reason why. Those tracks are just kicking right now.

4. It knows what you hate

The algorithm is watching what you skip, and learning to avoid those songs or artists. So if you’ve got a favourite genre but one artist within it you can’t bear, it’s unlikely to appear in your playlist at all (or at least, not for long).

5. It can spot unusual patterns and rationalise them

Finally, Spotify is wary not to pay too much attention to small forays outside of your normal patterns. So you won’t get stacks of kid’s music popping up if you happen to play a track for your toddler. Or if your mum spends Christmas day playing Cliff Richard on your account, when that’s far from your normal jam, it knows to avoid.

(The only downside of this feature is that if you suddenly get into gangster rap when you’ve only previously listened to The Carpenters, it might take a while for the algorithm to catch up and serve you some DMX. Be patient, that shit is worth waiting for.)

These insights are just the tip of a very complicated and impressive technical iceberg. One that’s being constantly shifted and refined to provide the best possible service and keep Spotify ahead of competitors.

However.

Does it work for everyone?

Despite being awestruck by Spotify selecting that Madonna track, my husband saying he is always finding new artists through it, and the entire world bowing to the superiority/machines-are-taking-over terror of the Discover Weekly recipe, I’ve never felt the need to listen to my own list with any regularity.

Occasionally I’ll read through the suggestions, literally say “meh”, and go listen to something else. I know, right! Not that magical now, huh, Spotify algorithm-worker-nerds (PS surely the greatest job ever).

I’ve always thought it was because the technology works best when you use the platform regularly. I’ll occasionally listen to it on work days or at the gym, but mostly I put the radio on or have good old silence.

Sometimes, but only sometimes, in moments of rare egotistical abandon, I’ve thought it’s because my music taste is SO GOOD it cannot be captured by a machine.

I KNOW.

I suppose I’d better take another look just to be on the safe side.

Inside ‘Discover Weekly’

I take a look at my list. As usual there are lots of artists I’ve heard of but don’t feel particularly excited by, and songs by artists that I like but I’ve already heard plenty. Plus Elbow, who I fundamentally object to.

At this point I would normally make a noise of contempt, then flip over to my own playlist or turn on the radio.

But maybe I’ll put it on this once. Just for science.

Hmm, maybe China Crisis are more my thing than I thought.

And no, I hadn’t heard that Wilco song, yes, and it is very pleasant.

Mmm Jane Weaver, that’s rather lovely isn’t it.

Well, this is embarrassing. Turns out not only is this thing smarter than I thought, it appears to have selected music that I DIDN’T EVEN THINK I WOULD LIKE. But did.

Spotify actually knows me better than I know myself. Are you fearing for your jobs yet, friends? You should.

Because AI has arrived, and it is telling me to listen to Blancmange.

(Please note that I still think Elbow suck).

This post is part of the #write52 project, the brainchild of Ed Callow. He even let me guest edit the newsletter last week and it was INTENSE. You can find out more and sign up to the newsletter on the #write52 website.

Hi there! I’m Penny, a freelance copywriter. I decided to write about a different number one single each week for my #write52 project. Just to be absolutely clear, not just any random number one single, but a song that was UK number one IN THIS VERY WEEK. If you look up the dictionary definition of a fool’s errand, this will be in there.

If you want to find me in social media life, you can follow me on Twitter, where I keep trying and failing to avoid politics chat, and Instagram where I post long captions under pictures they bear no relation to.