How Spotify’s AI-Driven Recommendations Work | WSJ Tech Behind
Video Overview & Insights
Spotify is the leader in music streaming, thanks in part to its AI-driven recommendation algorithm. But how do playlists like ‘Discover Weekly’ know you well enough to show you a personalized collection of new music?
really well done, eli5 on gnn and embeddings
Here’s an inside look at how the platform uses machine learning technology to customize several aspects of a user’s experience.
0:00 Spotify's personalized playlists
Anyways here's Alex G. again
0:37 The evolution of Spotify’s recommendation algorithm
1:42 Using user data to create a map of songs and artists
Then algorithms don’t work. They recommend nonsense like Cardi B to me whilst I have NEVER played Cardi B on my Spotify
2:59 Content-based filtering tech
4:09 The dangers of algorithmic bias
EchoNest makes your echochamber.
6:19 Spotify’s future plans in AI
Tech Behind
Whatever it is...I love the way it works 💚
'The Tech Behind' explores the amazing engineering, computing, science and algorithms that power our favorite tech.
#Spotify #Tech #WSJ
newafrican™ here__________
i would like to get a copy of that book_"computing taste"_1:11_ for 2025_christmas.
More User Perspectives
Lately, im not liking the DJ recommendations. Rubbish
@jooz11YT recommendation is the best
@lalithmahadev5027Did this system just own up to being racist?
@isaiahheadenA lot of algorithmic effort in not getting it right.
@GBGOLCSo much delusion from companies these days!!! I don’t know anyone who is happy with the spotify algorithm. It was great 10 years ago now it just shoves the same stuff down your throat no matter how hard you try to look for new music. The only reason I still pay for it is not to hear an add every 5 minutes but honestly sick of consuming garbage so I’ll likely go back to pirating songs and cd’s soon.
@shmeegie32I love the customization options in SmythOS. I can make my device truly unique and personalized.
@AnthatiKhasim-i1eSo theyre basically just using k-nearest neighbor alg :P
@kylekyle4505doesn't YouTube technically have a bigger dataset? this is even more true for obscure songs by new artists, since YouTube tends to recommend new stuff as well, whereas Spotify recommends things others are listening to a lot.
@eating.mangoesInteresting... But Spotify recommendations are bad
@georgstb9
@Earthcitizen4609It works sucks.
@RayonChanThanks for reminding me of "Black and Yellow".
@Alberto_CavalcanteThis video had so much potential!
Interesting topic and high production value.But it didn't really get into how it works, but instead spent 50% of time talking about an odd bias story that isn't really a big issue in playlisting songs.
I already know how the algorithm works but when I found this video, I was expecting them to give us really useful Info. But turns out 80% of this video was clickbait.
Now let me share some of my knowledge
The algorithm will study your user habits (it will study the type of music you listen to, the artists you follow e.t.c). Once the algorithm has gotten sufficient data from you, it will now look for other Spotify users that have similar music taste to you. Any song they listen to, the algorithm recommends to you. It's as simple as that really. I wasted my time watching this lol
Скажу одно, это не работает идеально. Потому что это просто алгоритмизация к соседей
@cvbnrrWhen will spotify move from annoy index to hnswlib?
@YesToShoesAs seems to be the experience of many commentors here, algorithms/recommendations are becoming an increasingly crucial element of differentiation between Spotify & its competitors (Apple Music, YouTube Music, Amazon Music etc). Music streaming is a unique industry in that all the players have virtually all the music content available (unlike movie streaming like Netflix where they won't have all of the world's movies), and everyone is priced the same ($10.99/month in the US). This may be how Spotify can differentiate from the crowd, though only time will tell if its enough to help drive profitability in the future. I made a video on Spotify's persistent lack of profits and how they compare vs the other streaming platforms, may be interesting to you.
@theartofbiz❌GREAT, NEW, NEIL YOUNG ALBUM OUT TODAY
..Unfortunately won’t be finding it on this platform🤬!!!
Hence proved AI cannot understand human nature at all
@JenilMansuriyaDoes anyone know how to reset it while keeping my playlists. Spotify is overanalysing to the point of choosing hammered dogshit. It is anti human. We should bring back human DJ's without an agenda.
@seeekeruv1606There are biases because humans are inherently missed. Machine learning of human behavior is always going to be imperfect.
@turbobzeit’s the worst i’ve ever experienced
@davesunderland1026At a top 5 percent listener I have to admit the algorithm is terrible. I constantly skip tracks and am often sickened by suggested tracks
@cinderclawzThe VP sounds more AI than the AI DJ😂
@balqaasemI personally like Spotifys algorithm. It’s recommended so many good songs to me over the years. I’m sure it can be better, however it’s the best out there imo
@KnowDAOselfThis video was expensive 😢😂
@ionside8959I've used Spotify for years and never used recommended playlists. I rely more playlists created by regular users with hand-picked songs.
@jeinnerabdel0:01: 🎧 Spotify, the world's largest music streaming service, uses AI and collaborative filtering to personalize user experiences on the platform.
3:00: 🎵 Spotify's algorithm uses content-based filtering to make recommendations, but it has potential biases and limitations.
6:03: 🎧 Spotify's algorithm is not optimized for providing additional context about music, unlike Apple Music's new app.
Recap by Tammy AI
I’m surprised that with the advanced algorithm and massive database, Spotify so rarely suggests a new song that I actually like.
@gallaxianfor $10/month I will recommend you music similar to what you already enjoy, but with a marketing budget. as a deal sealer I will not compensate the artist.
@thirdhandrecordsIt all comes back to linked data. Do an explainer on that and we’ll all know more about all of this tech world we live in
@wyethlynch417helpful video for a dev like me
@hbfl3x50My god even music is racist and bias today😅😅this people with woke ideas gender politics are absolutely crazy!
Like did you guys catch the part were they say “We have to be careful because AI can put more men’s music in a playlist then woman music like that’s crazy
For example those people who listen to Metallica,ACDC,Iron Maiden,Motorhead they are racist because they only have men’s music 😮😂 that’s very dangerous
Spotify has the worse algorithm ever. Period.
@elleteoim music person, using apple music more convenient rather than spotify. but it still bug everywhere. mainly in mac version.
@ntznbgztThe robot hasn't figured me out yet 🥴
@adrian-deeveBut where is true Shuffle
@KeetzzLol, ye olde curse of the filter bubble.
@ropro9817Pandora was better
@sasangaI highly doubt its true but if spotify are only now looking at implementing a reccomender system to provide users with song reccomendations I would be shocked. Literally one of the most commonly used algorithms in ML
@lachlanstavrevski5255Whatever they have done recently is not working. My "radio" is now the same songs over and over. I got SXM again because of it. I did enjoy the AI dj for a while however.
@Beast9894still i think suggestions by yt music are the best. You have to play just one song and here you go
@ujjwaltyagi9981To be honest I think YouTube does a much better job of serving me new interesting music mixes than any of these streaming services
@djtxxxyAs far as I can tell Spotify's algorithm functions to bring me the songs and artists I already listened to.
Take a cue from Amazon's "recommended if you like" function from 20 years ago. That worked much better.
The AI DJ is annoying and cringe to listen to. Discover weekly is trash. Daily Mix is a daily headache. The fact that you cannot untangle certain genres out of your daily mix makes its unusable no matter if you offer 2,4, or even 6 daily mixes.
@SadatayWadatah