2 ms·
It seems like there is some pitfall in making a recommendation algorithm that more and more people are falling in to. I don't watch TV or have Netflix, but I l
by perl4ever 7y ago
It seems like there is some pitfall in making a recommendation algorithm that more and more people are falling in to.
I don't watch TV or have Netflix, but I listen to streaming music, and I noticed something weird. I used to listen to Pandora, and it recommended a lot of different artists that I liked on hearing them for the first time. It single-handedly gave me a taste for 21st century music. But there wasn't enough music in their catalog, so I switched to Spotify. Now Spotify will play more of the songs by people I already like, but very nearly everything it plays that I didn't find on Pandora I hate and it keep playing the same things over and over no matter how quickly and frequently I skip past them.
It's hard for me to believe that they could be this bad and unresponsive unwittingly, but I wonder if perhaps there is some kind of naive algorithm that is failing.
...you may say, why don't I explicitly mark songs "like" or "dislike", but I mostly listen in my car where I can't do that.
One alternative hypothesis is that different people use Pandora vs. Spotify and the ones on Pandora have better taste, but I can't believe that's all of it.
- jdsully 7y agoIts less risky to give you things you already like than to try and guess if something new will be appreciated. Its the same reason hollywood keeps doing remakes. If you only go by statistics and don’t think through the “what if everyone did this” question then giving you the same old content on repeat will have better metrics.
- zxcmx 7y agoRight, it feels like things are optimising purely for “next click”, i.e, what will keep you mindlessly in flow, forgetting the time and consuming more content. So if I’m watching a video about TIG welding say, to optimise “next click” I suppose I should be recommended mostly more welding videos, with a smattering of engineering and finally a bit of clickbait mixed in case I’m starting to get distracted. If services are optimising for “keep going right now”, they might be modelling as something like a markov chain of “clicked on next” and that’s how we get stuck in this kind of short-sighted local optimum.
- codingslave 7y agoYoutube does this too, and its really killing the recommendations for me. Same thing with Quora, I used to browse Quaora all the time, until the content ended up just being the same. I think in the aggregate these algorithms are driving more engagement, and pissing off some people, but not enough for them to change the product.
- Osiris 7y agoPandora has almost no variety at all. Even a playlist like "90s Alternative" which should have thousands of potential top songs plays the same songs every time I listen Google Play Music has been the best for me in terms of variety, but it's buggy as hell. It freezes up if it can't play an ad for some reason, won't automatically play when bluetooth is connected (like Pandora does), and other annoying things. (I ride a motorcycle so I can't mess with my phone if something goes wrong)
- onion2k 7y agoPandora has almost no variety at all. Even a playlist like "90s Alternative" which should have thousands of potential top songs plays the same songs every time I listen If a streaming service didn't order automatically generated genre playlists by popularity (with a little randomness) people would stop listening to them and start saying the playlists are full of rubbish songs they've never heard of.
- perl4ever 7y agoI don't think I started with a playlist like that. It's been a while, but I think I started with specific artists/music (albeit from the 90s) that I liked. I don't think you can expect individually satisfying suggestions if you request generic music in the first place.