I have pretty much exhausted all my usual sources of music, so I thought I would train YouTube's recommendations to recommend me good music. From the videos it recommends me I know firsthand that it's pretty good at recommending, so I wanted to try the same with music. I also tried quite a number of other music recommendation sites, but they weren't that good, I believe YouTube can do it better if I train it right.

The best idea I had so far for this was to put music I like into a playlist, duplicate especially good music (duplicating entries is surprisingly difficult) and let it play on shuffle repeat with double speed on my Raspberry Pi 3. Sadly even with an audio-only Chrome extension, it still freezes very often, because the Raspberry Pi just doesn't have enough power, it's not made to stream videos.

So my next idea was to just open about 100 tabs at once on my pretty powerful laptop and let them all play audio-only double-speed shuffled repeated music. I haven't yet tried that, it should work (maybe not with such a high number), but it's a pretty ridiculous thing to do.

Isn't there an easier way to tell YouTube that I've watched a video? Maybe an API call that tells YouTube that I've watched a video, without actually having to load and play it? Or something opposite of the "not interested" button on recommendations? Or maybe something with Google Music (don't know very much about that or if it's even related to YouTube)? Or do I actually have to loop the music I like until I also like the recommendations?

Also posted here.

  • I just found something that might work, but it probably also depends on my watch history as well, so it likely doesn't solve my problem completely: music.youtube.com/library->click on three dots of a playlist filled with music->"start radio". Seems to recommend music based on that playlist. – Fabian Röling Jun 29 '18 at 12:22
  • Since I asked this, I used the YouTube mix I get when searching "music" very often, that gives at least mediocre results usually, but I have to remove many tracks which I've heard before, which aren't even music, which are played live or that I obviously don't like (e.g. pop, hiphop, ...). And even then the recommendations aren't nearly as good as I hoped and aren't really too related to my liked/favourited music (which I have now added, after cleaning up both a lot). I thought an AI that is this good in recommending me videos would learn faster what music I like, but apparently not. – Fabian Röling Oct 17 '18 at 10:06
  • Technically an answer to my question is using youtube-dl to mark videos as watched and skip the actual download. But sadly it doesn't help: I actually did the "play as loop in many tabs" and since then I've also played a lot of the music mix and consistently removed everything I didn't like. For a while I got quite alright recommendations, but then apparently something changed internally and now it's bad again. So I would still like to know how to train the YouTube AI better. – Fabian Röling Sep 13 '19 at 20:54

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.