Continuing our work to improve recommendations on YouTube • Hacker News Board

Re: Continuing our work to improve recommendations on YouTube

#171

Post by checkyoursudo » Fri, Jan 25, 2019, 9:21 PM UTC

reaperducer wrote: post #157

TeMPOraL wrote: post #136

Earlier quoted context omitted.

Tangent: I'm expecting a kid, and time will come in couple years, when I'll want to show them kid videos. I fully intend to set this up as streaming from a library curated by my wife & me, populated with youtube-dl and through other means. I know some people on HN have such setups; could you share recommendations for hardware and software stack? I suspect a Raspberry Pi may not be enough (having network and USB shari…

If you're in the United States, 90% of the population can receive the 24/7 PBS Kids channel over the air for free. If you're worried about the quality of content from YouTube, or the chances that someone gaming the system dumps questionable content into your stream, it's worth spending ten bucks on an antenna. It's also good for supplementing your child's YouTube viewing when YT starts showing the same stuff over and…

The PBS kids website has a lot on it as well, all ad-free.

Also PBS kids apps, ad-free as well.

Re: Continuing our work to improve recommendations on YouTube

#172

Post by TeMPOraL » Fri, Jan 25, 2019, 9:24 PM UTC

reaperducer wrote: post #157

TeMPOraL wrote: post #136

Earlier quoted context omitted.

Tangent: I'm expecting a kid, and time will come in couple years, when I'll want to show them kid videos. I fully intend to set this up as streaming from a library curated by my wife & me, populated with youtube-dl and through other means. I know some people on HN have such setups; could you share recommendations for hardware and software stack? I suspect a Raspberry Pi may not be enough (having network and USB shari…

If you're in the United States, 90% of the population can receive the 24/7 PBS Kids channel over the air for free. If you're worried about the quality of content from YouTube, or the chances that someone gaming the system dumps questionable content into your stream, it's worth spending ten bucks on an antenna. It's also good for supplementing your child's YouTube viewing when YT starts showing the same stuff over and…

Not in the US; unfortunately, I don't think Poland has a dedicated government channel for kids.

Still, I want to make it video-on-demand, but with content fully curated by us - to avoid age-inappropriate content, ads and recommendations. There's no way I'm going to let my kid touch raw YouTube in the next decade.

Re: Continuing our work to improve recommendations on YouTube

#173

Post by TeMPOraL » Fri, Jan 25, 2019, 9:25 PM UTC

explainplease wrote: post #166

TeMPOraL wrote: post #136

Earlier quoted context omitted.

Tangent: I'm expecting a kid, and time will come in couple years, when I'll want to show them kid videos. I fully intend to set this up as streaming from a library curated by my wife & me, populated with youtube-dl and through other means. I know some people on HN have such setups; could you share recommendations for hardware and software stack? I suspect a Raspberry Pi may not be enough (having network and USB shari…

It may not be much help, but I've found minidlna to be a reliable, simple, lightweight media server on my LAN. I expect it could run well on a low-power device. Any DLNA/UPNP device can stream from it.

Thanks, I noted it down and will check it out.

Re: Continuing our work to improve recommendations on YouTube

#174

Post by sayno3 » Fri, Jan 25, 2019, 9:28 PM UTC

How about not allowing for using ad campaigns for targeting individuals or even groups of individuals by email or other personal information, you fucking retards (youtube employees responsible)? How does it feel to know that your "don't be evil" corporation is what allows for the radicalization of fringe/psychologically-vulnerable users? They have literally placed a greater importance on fine-grained targeted advertisement than one of the most subtle forms of malicious manipulation that has probably had a great deal of influence on the worst atrocities of our time.

Re: Continuing our work to improve recommendations on YouTube

#175

Post by sayno3 » Fri, Jan 25, 2019, 9:32 PM UTC

What needs to be in recommended videos is niche stuff that falls into similar categories of things you have watched. That is what needs to be recommended, not a video with a million views already which compounds the winner-take-all tendency of the internet and memetic information spread. Something that is so hugely popular will be recommended through word-of-mouth or posts from other sites. They are making it harder to find lesser known pieces of information, essentially burying valuable stuff in a torrent of already popular videos.

Re: Continuing our work to improve recommendations on YouTube

#176

Post by magicalhippo » Fri, Jan 25, 2019, 9:38 PM UTC

> We now pull in recommendations from a wider set of topics—on any given day, more than 200 million videos are recommended on the homepage alone.

What does the first part (wider set of topics) have to do with the second part (number of videos recommended)?

Anyway, YouTube recommendations are generally quite poor I find, though sometimes a few gems come along. I also made the mistake of watching YouTube on my Android TV box. Holy cow, an ad _every single minute_. And the same 4 ads, rotating.

Re: Continuing our work to improve recommendations on YouTube

#177

Post by throwawaysea » Fri, Jan 25, 2019, 9:40 PM UTC

jlawson wrote: post #51

bengale wrote: post #3

It does seem to be getting better, for a long while I had to make sure to clear my history of any conservative leaning videos or I'd end up down a rabbit hole. Too many and all of a sudden its bloody flat earth videos and other insane conspiracies. Made it very difficult to try and watch a balanced set of videos.

This doesn't match reality. I've watched lots of conservative videos and I've never seen anything about flat Earth or "insane conspiracy theories". Honestly this comment thread seems like trying to build a narrative bridge between straightforward right-leaning content and bizarre conspiracies. The goal is to use censorship of bizarre conspiracies to justify censoring right-leaning content by conflating the two. I'm v…

This is exactly my experience. I've watched many videos from conservative content creators on YouTube to understand their perspectives, and by and large I found them to be reasonable and not dissimilar in their balance to the content I usually watch. I did not find conspiracy theories, flat out lies, or other extremist content either nestled within those videos or recommended to me in other videos. They may exist out there, but I haven't been exposed to them.

The Buzzfeed article (and others like it) is ultimately just a collection of anecdotes. It is not data that reflects my reality and I expect it does not reflect reality for many of YouTube's users. However, it provides a lot of confirmation bias for those who look at the existence of content/perspectives they disagree with and have an urge to want to erase that content wholesale. It feeds into what seems to be a crisis manufactured by some activists on Twitter and some news outlets (like Buzzfeed).

And of course, it dehumanizes the right and their experiences/perspectives by associating them with terms like 'radicalization', or 'conspiracy theory', or 'fake news', which are probably not only exceedingly rare but also present on the left.

Re: Continuing our work to improve recommendations on YouTube

#178

Post by maxxxxx » Fri, Jan 25, 2019, 9:42 PM UTC

rapind wrote: post #170

mindcrime wrote: post #145

My biggest problem with YT recommendations is that they dramatically overweight recent videos when generating the recommendations. I mean, I can spend a week doing nothing but watching (well, listening) to videos of classical music performances, and my recommendations will be full of classical music, exactly as you'd expect. Then I watch one random one-off thing, like "15 funny pitbull fails" and suddenly ALL of my r…

I have the opposite problem. I'm only interested in new videos for the most part, and my recommendations are filled with videos from months ago. It would be much better if they would give us a flag or some date range options. Obviously this increases UI complexity, but maybe it doesn't need to be that prominent?

"Obviously this increases UI complexity"

It always annoys me that giving users options they can control is viewed as complex but search engines that return whatever they deem as important aren't viewed as complex.

Re: Continuing our work to improve recommendations on YouTube

#179

Post by thrusong » Fri, Jan 25, 2019, 9:43 PM UTC

NoodleIncident wrote: post #168

thrusong wrote: post #68

Why do they recommend the same five videos of a series (like Barefoot Contessa) over-and-over when I've watched them already? I don't want to say "not interested" because I don't know what that means... I want a feature where it's like "more like this, but not this one in particular." It's surprising they've been under Google's control for a very long time and yet their main recommendation algorithm still leaves a lo…

If I remember correctly, the "not interested" UI has an option for "I've already seen the video"

That just makes it seem like their algorithm is powered by bad code or bad management, or both. They know what I've watched. They should know if a video is likely to be watched repeatedly (ie. classifying songs/music videos differently from other content).

Re: Continuing our work to improve recommendations on YouTube

#180

Post by AnIdiotOnTheNet » Fri, Jan 25, 2019, 9:43 PM UTC

NoodleIncident wrote: post #168

thrusong wrote: post #68

Why do they recommend the same five videos of a series (like Barefoot Contessa) over-and-over when I've watched them already? I don't want to say "not interested" because I don't know what that means... I want a feature where it's like "more like this, but not this one in particular." It's surprising they've been under Google's control for a very long time and yet their main recommendation algorithm still leaves a lo…

If I remember correctly, the "not interested" UI has an option for "I've already seen the video"

Ok, but why? If there is one thing the algorithm should know it's which videos you've already watched!