2 ms·
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
by imglorp 6y ago
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 http://invidio.us, code at https://github.com/omarroth/invidious https://github.com/omarroth/invidious, which would be a nice starting point.
- someguyorother 6y agoYou 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 http://www.rongpan.net/publications/pan-oneclasscf.pdf