Apparently some previously-scrapped data for older videos that gets updated, and heuristic estimates combined with extrapolating data from those who use that extension and like/dislike. LTT did a video where they found the accuracy really good.
Beating Google at their own game, they (apparently) are.
I think the accuracy doesn't even matter too much. Personally, I believe I base my decision to watch the video after clicking on it on the ballpark the like/dislike ratio is in. "Almost only likes", "2/3 likes", "50/50 likes/dislikes", ... For that purpose, the extension does its job.
> with extrapolating data from those who use that extension and like/dislike.
Often you see the opposite, where a third party service provides customers with a feature extending the features of a big service like youtube or twitter, and then the big service buys the startup offering that extension, or integrates the feature themselves, leaving the extension dead. Now we see it happening in reverse direction, very interesting.
Apparently, the Youtube API documented here - https://developers.google.com/youtube/v3/docs/videos - returns both likes and dislikes. All Google did was remove the rendering of that API response field onto the webpage.
Extensions probably acquire the video url and make their own API call just to retrieve the dislike count. To me, that seems cleaner than trying to snoop on the main API call response (I think extensions might not have access to those, maybe they do). Basically, the data is available via API calls.
That's the analytics API. We already know they show dislikes to the channel owner. Isn't this the same thing?
And your other comment says "The dislike count is also still present in the [0] video API response." but the page says right there "Note: The statistics.dislikeCount property was made private as of December 13, 2021."
https://chrome.google.com/webstore/detail/return-youtube-dis... (also available for Firefox, Edge, ..)
Of course, it doesn't solve the recommendation issue the thread creator mentioned.