Hacker News Reader

Comment by distances | Hacker News Reader

I clear my YouTube search and watch history about once a week. Partly because of privacy, but also because a single binge of, say, metal casting videos does not mean I want them recommend ever again in the future.

View on HN

One thing this opens you up to is the average recommendation, which is biased towards inflammatory or click-bait content.

I do it every week as well and the recommendations start off being seeded from my subscriptions, not from the trending tab.

They use browser fingerprinting and/or IP addresses as well. Even on a brand new browser session, doing something even slightly related to the previous session brings back its entire recommendation history.

Yeah, the fact that I can clear cookies, open a private browsing window with tracking protection turned on, go to YouTube, and be asked which of my two gmail accounts I want to log in with, is pretty creepy.

> doing something even slightly related to the previous session brings back its entire recommendation history.

Are you sure that it does, and that it's not just a case of "hey we've never seen this person before, but they watched X, let's immediately start with recommendations Y and Z because other people who watched X were engaged with it?"

I used to think that and gave them the benefit of the doubt, but then I realised that some of the videos being recommended had nothing to do with the one currently watched other than the fact I watched similar ones previously.

How does that work at, say, the library, or some other public internet kiosk? Or maybe they're just assuming these are edge cases today, amidst the billions of private browser instances on handhelds?

It probably doesn’t work, considering the library users would be watching totally different & unrelated videos, where as in my case they have years of data of very specific viewing habits centred around a few topics.

Yes you are right they are using browser fingerprinting, but with the right tool they can easily change it. They can use Kameleo software or Multilogin etc. https://kameleo.io/

The problem with this is that it then just suggests the content that hits the front page instead - so for the most part a load of crap. All I'm doing is swapping recommended videos on, say, metal casting, for YouTube's "on-brand" content creators which pump out generic content on a bi-daily basis at 10 minutes in length.

Reply on news.ycombinator.com