Hacker News new | ask | show | jobs
by dvlsg 1732 days ago
YouTube still recommends stuff I hate all the time, despite me explicitly trying to tell the system I'm not interested.

To pick an example, my YouTube recommended section is full of videos of people reacting to the things I actually like to watch. I don't want to watch people reacting to the things I like. I just want to watch the things I like. I try to tell YouTube that by saying "Not Interested" to all of the reaction videos, but it doesn't seem to make a difference.

3 comments

I use uBlock to stop this type of shenaningans. I set filters for specific channels or "channel networks", for instance if I don't want to see any Vice related video show up in my recomendations, i.e. "Vice Fightland", "Vice bla bla bla" etc, I set a rule such as:

www.youtube.com###dismissible:has-text(/Vice/)

This way, any channel that has the word Vice in it will never show up to me in recommendations, be it on the main page or the sidebar. This "has-text" regex is really useful, not just on Youtube. I use it a lot, all over the internet. In your case, switch "Vice" with "react" and I believe you'll get rid of 99% of these videos showing up on your feed.

It'll also get you rid of 99% of React.js tutorials.

A has-text condition always has this risk of matching too much.

Those types of videos still tick the box of "relevant" to current recommendation algorithms though. I sort of mean that algorithms won't branch into unfamiliar territory. So, as an example, I've never looked up "underwater basket weaving"[1] or getting into to. So from the algorithm's perspective, there is an element of uncertainty with recommending it, because it doesn't know if I'll like it, be ambivalent, or strongly dislike it. Then, it compares that rating to other content on the platform, like your reaction videos, and even though they may be disliked, if they are favorable to others in your focus group, they'd still have a higher rating than uncertainty.

The issue is that we don't really have a good method for measuring uncertain interests and since most algorithms have an underlining "make the company money", we don't get anything beyond the same ol' same ol' recommendations.

[1] Underwater Basket Weaving is a placeholder for anything - yoga, video games, cooking, politics, essentially anything you are currently NOT getting presented.

I loathe videos on trending with the exception of some science content. I say “not interested” all the time but the third recommendation is always some bs I have zero interest in.