Why is everybody pretending the recommendation system is designed to show you what you want to see, rather than to maximise ad revenue?
I assume this because I pay and don't see ads.
YouTube’s ‘dislike’ barely works according to new study on recommendations
91–100 of 437 posts
Re: YouTube’s ‘dislike’ barely works according to new study on recommendations
#92Every other day YouTube seems to want to give me another American local news channel, despite the fact I'm in Australia. I've marked them as "don't recommend" as I've seen them, but I'm still getting a lot of results for Fox Local Clapistan 7 all the time which just aren't relevant to me at all.
Re: YouTube’s ‘dislike’ barely works according to new study on recommendations
#93The trick to improving YouTube recommendations is to regularly prune uninteresting videos and channels. Eventually this should start surfacing videos you're more likely to find interesting. I've actually hit the end of my recommendation stream on the mobile app after marking dozens of videos with Not Interested. Once this happens it starts recommending videos you've already watched or older videos from creators you'v…
>Some might say that you can still use search At some point Youtube search was neutered. I get 5-10 results for any given query before Youtube gives up and just offers me recommendations rather than search results. Interestingly, if I search google for [search term] site:youtube.com, I actually get pages of results which are never present in the Youtube search. Sadly, it seems that Google has intentionally diminished…
Yup. Search reveals two or three relevant videos but from that point on the search results are indistinguishable from the regular feed. And I get the same video recommended to me all the time. Sometimes back to back!
Re: YouTube’s ‘dislike’ barely works according to new study on recommendations
#94The trick to improving YouTube recommendations is to regularly prune uninteresting videos and channels. Eventually this should start surfacing videos you're more likely to find interesting. I've actually hit the end of my recommendation stream on the mobile app after marking dozens of videos with Not Interested. Once this happens it starts recommending videos you've already watched or older videos from creators you'v…
> Some might say that you can still use search, but that has been crippled to include unrelated recommendations after one page of results. Not only that, I want YouTube to help me discover new content. When the recommendation system works and it shows me new kinds of interesting stuff, it's superb. TikTok is eating YouTube's lunch in this regard. this and it's going too far, since Youtube is now polluting people's se…
Google have a ton of other ML features about you that they fit to the model and so you have been profiled to be a bit lecherous.[1] Do you feel good about that?
Think how it must feel to be profiled to be a crook or a bad credit risk or something else that actually impacts your life and is false. It's a little worrying.
[1] Other people like you like this so it is assumed you do also. Without "lecherous" ever being mentioned or even anything like suggestively dressed appearing as a model feature. It is really, really easy to accidentally train a statistical model to be wildly prejudiced. It seems significantly harder for us to categorize it when it happens as "other similar decisions are made by shockingly violent racist and evil humans and this model and those using it should be treated with the same or more derision as they claim not be so and ought to know better." Significantly harder for many good reasons as well as bad ones.
Re: YouTube’s ‘dislike’ barely works according to new study on recommendations
#95Why is everybody pretending the recommendation system is designed to show you what you want to see, rather than to maximise ad revenue?
I assume this because I pay and don't see ads.
You are in a minority. Why do you think they built their recommendation system with you in mind at all? You've already paid, you no longer impact the fit of the short term revenue maximization objective function.
Re: YouTube’s ‘dislike’ barely works according to new study on recommendations
#96Honestly, I feel like YouTube recommendation is the _only_ good thing YouTube has lol. I often see a ton of complaints about the recommendations but I've honestly never had an issue. I know that yes, you click on one video and now all the sudden you get 100s of that type... but it sorta works for me since it seems like that "effect" wears off just as fast as it starts.
For example, I have a friend that's into cars. I watched a few car videos and then 50% of my recommendations were car videos. It works for me though, because usually those videos are genuinely interesting. I haven't watched one in weeks tho, so going to my home I see 0 car related videos.
I will 100% believe everyone though when they say it just doesn't work for them. Just wanted to share my experience.
Re: YouTube’s ‘dislike’ barely works according to new study on recommendations
#97The trick to improving YouTube recommendations is to regularly prune uninteresting videos and channels. Eventually this should start surfacing videos you're more likely to find interesting. I've actually hit the end of my recommendation stream on the mobile app after marking dozens of videos with Not Interested. Once this happens it starts recommending videos you've already watched or older videos from creators you'v…
The biggest problem with this approach is that it takes a lot of curation to get your recommendation in a decent shape and it takes you one accidental video view to have your feed be full of uninteresting content related to that video. This asymmetry is the root of all the recommendation system problems.
For me as soon as I say "not interested" it completely nukes that keyword. I don't have to keep doing it.
Re: YouTube’s ‘dislike’ barely works according to new study on recommendations
#98Earlier quoted context omitted.
Wouldn't a good skill based matchmaking system take care of that?
That’s the problem. this player was at the top of the matchmaking ladder that (very small) population had a lot of people who wanted to avoid him he couldn’t get games at his current level, and he couldn’t rise above that level, because there was nothing above that level
Maybe that means he won the game and it's time to do something else until someone with similar skill comes along? Or maybe there should be a handicapping system.
Re: YouTube’s ‘dislike’ barely works according to new study on recommendations
#99Earlier quoted context omitted.
That was my experience years ago when I last did so. Logging in offered no visible upsides and multiple downs. I learned not to log in and use search wherever possible. Now I use Invidious (with JS disabled and/or uMatrix blocking YouTube functionality in the browser as an additional nudge), or youtube-dl / ytdl / mpv to play video directly. Usually audio-only for talkpieces. I'll also disable recommendations and com…
If you actually subscribe to channels you like when logged in, you'll get much better recommendations. I have over a 100. (And no, all of the recommendations aren't from my subscribed channels, it's a mix of them which have newer videos and similar videos to my subscriptions.)
I greatly prefer not feeding Google's data-hoarding dragon.
I've been banging this drum for a while:
https://news.ycombinator.com/item?id=6746731>
https://web.archive.org/web/20160307214624/https://plus.goog...>
That references a G+ post which is of course dead now with that service. Wayback claims to have captured the post but I can't find a version that renders properly (or at all):
https://web.archive.org/web/20131015000000*/https://plus.goo...>
(The HN submission referenced a post of my own on G+, which I believe I deleted shortly afterward.)
When I did utilise YT under a YouTube (and for a very brief time, Google+) account, I found the benefits de minimis. And in particular, what I'd most prefer would be to dismiss channels (or entire genres) with extreme prejudice, which simply was not available. And I'm not going to play rat to the Borg's Skinner box to find the reward morsel. I've got methods that work with ample sufficiency for me now.
So, no, not gonna bite.
Re: YouTube’s ‘dislike’ barely works according to new study on recommendations
#100Anyone else find it odd that we’re doing 3rd party studies on a programmed algorithm?