Manipulating the YouTube Algorithm [video] | Hacker News Reader

Manipulating the YouTube Algorithm [video]

youtube.com

116 pointsespeed38 comments

2 years back Brian Dean published statistics from analysing 1.3 Million YouTube videos[1].

Couple of key insights were,

Comment count strongly correlates with higher rankings

Average length of first page video is : 14 mins 50 secs

I would say, if someone starts a new Youtube Channel today with original content. Unless it is >10 mins (purposefully not being succinct), cautiously scripting the content to encourage comments (which are likely to be off-topic, polarising, memes, sarcasm etc.); then there's no way the video is going to be on first page (or) gain large subscribers.

[1]:https://backlinko.com/youtube-ranking-factors

Edit: Added /n for insights

Edit: I meant \n .

He just uploaded the second part of this series an hour ago: https://www.youtube.com/watch?v=V-1RhQ1uuQ4

This one is on Twitter, and in it he also talks with NATO, who contacted him after he posted the first video and sent him a NATO report on platform manipulation. See...

Robotrolling https://www.stratcomcoe.org/robotrolling-20183

In George Orwell's 1984, there is a description of a "kaliedescope" machine that remixes plot elements, etc from books to create media for the "proles." Basically, technology is at the point where we can do basically that for news stories. I wonder how far down the road TV shows and comic books could be? (I'm guessing 10 years, to get to the point where the first news synthesis by AI was a few years ago.)

Isn't this what most modern movies are, only the remixing isn't automated?

I wonder what a script for an action movie would look like if it was just based off of statistics of all the previous action movies. Probably watchable.

Isn't this what most modern movies are, only the remixing isn't automated?

That was part of George Orwell's point.

I wonder what a script for an action movie would look like if it was just based off of statistics of all the previous action movies. Probably watchable.

We could probably get some great comedy out of it!

Netflix is sort of already doing this, producing films based on your content and metrics of what you search for and watch. They aren't automating the actual movie bits yet, but it's progress.

> I wonder what a script for an action movie would look like if it was just based off of statistics of all the previous action movies...

Have you seen the Transformers movies?

> ...Probably watchable.

See above.

Brave New World had auto-generated pop-music which is essentially what modern pop-music is

Do you think Deep Learning can solve being charming and sexy? Maybe the general public will like auto-generated people better than the organic ones. (In this case, "organic" is technically correct, as it can be synonymous with "carbon based.")

EDIT: Come to think of it, just how hard would it be to use Deep Learning to create superhuman virtual Instagram models? I think we're closer to that than to self driving cars.

Come to think of it again, what about virtual Logan Pauls?

The world doesn't need any more Logan Pauls

The world has 10000's of them. The problem is that the quality control is off.

At more than one point in the video he says that the spread of this kind of automatically generated content was not YouTubes fault, but in a way I think it is. It seems likely that the methods of manipulation shown in the video aim more at getting such a piece of content going and less at sustaining it. If "the algorithms" goal is to maximize viewer retention and this kind of autogenerated content keeps some people glued to the screen, watching ads and writing comments, it's really doing an excellent job, it's just the goal itself that is unethical.

Edit: Typo

Quite.

"YouTube Executives Ignored Warnings, Letting Toxic Videos Run Rampant"

The company spent years chasing one business goal above others: “Engagement,” a measure of the views, time spent and interactions with online videos. Conversations with over twenty people who work at, or recently left, YouTube reveal a corporate leadership unable or unwilling to act on these internal alarms for fear of throttling engagement.

Wojcicki would “never put her fingers on the scale,” said one person who worked for her. “Her view was, ‘My job is to run the company, not deal with this.’” This person, like others who spoke to Bloomberg News, asked not to be identified because of a worry of retaliation....

https://www.bloomberg.com/news/features/2019-04-02/youtube-e...

Why can't all social network make 2 factor auth. and verify account with text messages. This should cut out lot of fake accounts. They can also randomly send out email verification when people try to upload stuff. It might make the process bit annoying but in long run help the platforms out.

Because you don't want to put hurdles between your users and posting a message. When you do, you're just creating incentive to move to another platform.

Youtube is like an election with a trillion billion candidates all competing for your attention, not 2

One of the obvious solutions for this is to charge hosting fees, something like users paying $2 per video upload or so.

I think the video misses one basic point that so far youtube tries to fight fakes with technology instead of just limiting the number of videos uploaded or letting human moderators view the videos -- which would undercut their market dominance.

It is a lot easier to say, oh hey, we've got this technology and people are bad at their biases, instead for example, we're subsidizing this venture to gain market dominance so everyone is able to upload whatever on youtube, but with even this subsidy we are unable to do appropriate content moderation.

Foums and mail lists 30 years ago had that feature.

> One of the obvious solutions for this is to charge hosting fees, something like users paying $2 per video upload or so.

That sounds like solving shoplifting by making customers pay an entrance fee. The majority of youtube uploaders are not the same 50 celebrities that rule the trending page and make thousands if not tens of thousands of dollars per upload.

I'm amazed that people who's uploads regularly stay below 10 do upload at all, I'd be even more amazed if they started to pay serious money just to upload their daily morning chat. Personally, I tend to care more for what happens beyond the carefully selected trending page. The trending page summons my inner Kaczynski, the rest calms him down. I'd hate to see those people stop.

I understand your criticism of my argument, that in this case only what is popular would be available but not what is niche and interesting.

I still make a point that there should be some barriers of entry with humans (moderators?) as gatekeeprers to prevent the spread of this this kind of misinformation, and that would cost more money than what is happening now, it is up to youtube how they choose to recover that cost.

Trying to do some clever AI on videos after the videos are uploaded is trying trying to close the barn doors after the horses are gone.

> I understand your criticism of my argument, that in this case only what is popular would be available but not what is niche and interesting.

That too, but my point was mostly that barriers would hinder the business. Even if people could afford to pay, they don't want to, and there will be a place next door where they don't have to. The problem will be solved insofar that the users migrate away, and take the problems with them - and the ad revenue ;)

> and that would cost more money than what is happening now, it is up to youtube how they choose to recover that cost.

Youtube _makes_ money off of this sort of manipulation, they have literally no interest in stopping it. They just want the PR problem to go away.

Likely far below $2/video, and probably with a free-service rate tier (several vids/mo, say).

Though I see other areas.

Does youtube really use a simple visual hashing method for content id?

I thought youtube uses deep learning for everything now?

YouTube uses multiple algos and signals, and yet spam still gets through. In some ways social media spamming techniques are reminiscent of the early days of the Web -- when the Web was overrun with "SEO" spam from junk sites attempting to manipulate the Google algo in the late 90s / early 00s. Google eventually solved the problem of Web and Gmail spam for the most part, but social media spam is a somewhat different beast.

HN has been having to deal with the in-authenticity problem too, maybe even moreso as of late. Solving it may require a concerted effort from all of us -- a vigilant watch by those who can see through -- those who not only take the time to flag inauthentic activity as they see it, but also educate others by calling it out and pulling back the curtain in a way that reveals its nature so more may become aware and learn to see it too. Discernment is a skill that can be learned when we are cognizant and vigilant and conscientious about paying attention.

Discernment is a skill that can be learned when we are cognizant and vigilant and conscientious about paying attention.

Google, YouTube, and much of the current web is very easily manipulated by small but sizable and highly motivated fractions of the population acting together. It greatly over-represents the opinions of the extremes, particularly the voices which generate the most outrage.

Fear is the mind-killer. But we never knew the extent to which this could reach until it was combined with viral outrage. (Monetized and politicized viral outrage, to be precise.)

https://www.youtube.com/watch?v=rE3j_RHkqJc

They definitely use a mix, the visual hash is just an easy one to demonstrate and was clearly the primary one used at some point in the past judging from the prevalence of simply reversed/mirrored pirated uploads years ago.

presumably the purpose of these vids is to make $ from the pre ads, not spread propaganda.

¿Por qué no los dos?

Destin's bias and hypocrisy makes it hard to watch his videos.

In what way is he biased in this video?

In the pro-usa way.

I thought Destin was actually manipulating the algorithm.

Well it wouldn't be hard to make the argument that he does, albeit not as programatically as his antagonists.

What his channel feed looks like: https://imgur.com/a/OqCIKlP

For what it's worth, he seems to be pretty open about that in the video

And further, even if he is attempting to "manipluate the algorithm" he really is just trying to maximize his views for his content, which is much higher quality content than the average youtube video.

One of the best youTube science videos ever!

https://www.youtube.com/watch?v=RtWbpyjJqrU

Reply on news.ycombinator.com