5 ms·

I wonder if they could look at groups of users who up vote "bad" videos (as marked by their own internal reviewers), see what else they up vote, cluster those v

I wonder if they could look at groups of users who up vote "bad" videos (as marked by their own internal reviewers), see what else they up vote, cluster those videos, and down promote them in their algorithm. IOW use the users who like "bad" videos to help find other "bad" videos via an algorithm.

twanvl 8y ago
That sounds easy to abuse. Just create some users, have them like obvious bad videos, then have them like videos of <insert competitor>.
- ses1984 8y ago
  Literally any algorithm that simple is easy to abuse, you need to integrate any algorithm with fraud detection and user trustworthiness assessment.
  - luckylion 8y ago
    Either Twitter figured that they don't care enough, or it's not that easy. Mass-flagging is still a thing they haven't solved, and what are you going to do? Kick out your users because they're trying to manipulate your powerful algorithm to do their bidding? There's easier ways to make your audience find a new platform.
    - ses1984 8y ago
      It's not that easy, the cost of doing it right is astronomical and most users don't care.
gmiller123456 8y ago
This really only works in a low volume situation. There's about 500,000 hours of video uploaded to Youtube every day, so employees would only be able to actually watch an insignificant fraction of that per day. Additionally, they're unlikely qualified to judge whether a video is actually good or bad. They can judge the sound quality, or videography, but unless you're well versed in the field the video is about, it's pretty much impossible to determine if a video is actually offering novel or good information. Additionally, the bots upvoting content are going to have several thousands, possibly hundreds of thousands of bots. So detecting the circle is going to be very difficult. Even more so if you consider a lot of legitimate users will have the same taste and watch a lot of the same videos.

twanvl 8y ago

That sounds easy to abuse. Just create some users, have them like obvious bad videos, then have them like videos of <insert competitor>.

ses1984 8y ago
Literally any algorithm that simple is easy to abuse, you need to integrate any algorithm with fraud detection and user trustworthiness assessment.
- luckylion 8y ago
  Either Twitter figured that they don't care enough, or it's not that easy. Mass-flagging is still a thing they haven't solved, and what are you going to do? Kick out your users because they're trying to manipulate your powerful algorithm to do their bidding? There's easier ways to make your audience find a new platform.
  - ses1984 8y ago
    It's not that easy, the cost of doing it right is astronomical and most users don't care.

gmiller123456 8y ago

This really only works in a low volume situation. There's about 500,000 hours of video uploaded to Youtube every day, so employees would only be able to actually watch an insignificant fraction of that per day. Additionally, they're unlikely qualified to judge whether a video is actually good or bad. They can judge the sound quality, or videography, but unless you're well versed in the field the video is about, it's pretty much impossible to determine if a video is actually offering novel or good information. Additionally, the bots upvoting content are going to have several thousands, possibly hundreds of thousands of bots. So detecting the circle is going to be very difficult. Even more so if you consider a lot of legitimate users will have the same taste and watch a lot of the same videos.