this doesn’t surprise me at all. I work at a consumer company and we use neural networks to personalize feeds. Thousands of features feed in and each member gets a different feed. It doesn’t take much / long to personalize.
What I am surprised by is that YouTube or others dont use as effectively. Neural networks are very well known. or, is there something “next level” that TikTok is using that isn’t well known?
Also worth calling out that Netflix also used to use personalized recommendation engines, but eventually found that “top in the US” won out, which I found fascinating. If one person loves action movies for example, wouldn’t they find a list of those more appealing than a generic list of top 10? I sometimes get curious about top 10 but rarely actually watch them myself
My gut instinct is that Netflix can't use the personalized algorithmic approach that TikTok succeeds with because Netflix has to pay dearly for its content, where TikTok does not. Because of the cost structure of licensing, Netflix cannot afford to have the diversity of content that TikTok's user-generated approach enables. A small content library makes it impossible to build a meaningfully personalized feed.
It's also much harder to tell why somebody liked a 2h movie, vs a 15 second clip. The 15 second clip only has so many properties, where as there's millions of reasons I might like one movie but not a similar one. Also, as far as Netflix, id rather see more diversity, and not see the same movie made over and over again. I would wager TikTok viewers are more willing to watch repetitive things play out.
This is not as difficult to understand as it might sound, Netflix can only run recommendation based on the series its audience has watched, and you can only watch so many series in a week. Whereas for 15 seconds clips, the engine get feed hundred if not thousands times more data it would on netflix data. so
What I am surprised by is that YouTube or others dont use as effectively. Neural networks are very well known. or, is there something “next level” that TikTok is using that isn’t well known?
Also worth calling out that Netflix also used to use personalized recommendation engines, but eventually found that “top in the US” won out, which I found fascinating. If one person loves action movies for example, wouldn’t they find a list of those more appealing than a generic list of top 10? I sometimes get curious about top 10 but rarely actually watch them myself