News for hackers. Stuff that matters.
Re:Evaluation and euthanization of a dangerous honeybee colo... by imglorp on Thursday June 25, 2020 @10:07AM (#23640909)
I think we got nerd sniped. The algorithm worked.

However I'm curious about how someone could write their own customized recommendation engine, if the Youtube APIs even support that. One thing on my to-play-with list is http://invidio.us, code at https://github.com/omarroth/invidious, which would be a nice starting point.

Re:Evaluation and euthanization of a dangerous honeybee colo... by someguyorother on Thursday June 25, 2020 @11:42AM (#23641955)
You basically need the visit and like graphs (which videos were liked/watched by what users), and I don't think the YouTube API will give you that.

If you had those, a good place to start would be the alternating least squares algorithm.

The visit/like graphs are "one-class", which means you don't know if user x hasn't watched video y because x disliked it or just didn't know of its existence. That makes it harder to apply ALS, but it can be done, e.g. http://www.rongpan.net/publications/pan-oneclasscf.pdf