How DO the recommendation algorithms work on the big social-media and video services?
Individuals and companies alike are always hungry for information on their inner workings, with the personalised feeds of services ranging from Spotify and YouTube to TikTok and Instagram so crucial to their businesses.
YouTube is the latest platform to promise a reveal, with a new video featuring YouTube Shorts product lead Todd Sherman. Like most examples of the ‘we’re going to explain our algorithm to you!’ genre, the explanation carefully only goes so far, so that it doesn’t enable mass-gaming of the feed.
Still, with YouTube Shorts an increasingly useful platform for musicians and the music industry, there is sure to be interest in what Sherman can reveal about its algorithm: including what counts as a ‘view’.
“Some platforms treat a view as the very first frame. Where, like, if you were to just flip through a feed, you would just be accruing lots of views. They’re basically impressions,” he said. Place your bets on which platform he was talking about…
“What we try and do with a view is have it encode for your intent of watching that thing, so that creators feel like the view has some meaningful threshold that the person decided to watch.”
The conversation also touched on video length (batted back with “what length do you need to tell your story?”); thumbnail design (not that important: “most thumbnails are never seen, because people are in the feed…”); whether hashtags matter (possibly); and whether YouTube optimises according to the time a video was posted (no).
Sherman also encouraged people not to try to hit some kind of quantity threshold in terms of how many videos they post to Shorts, but rather to make fewer, better clips. “There’s also something to be said about iterating: posting a video and learning from what worked or what didn’t.”
He also warned people not to delete videos and repost them to get more views. “I’ve heard people talk about this as like a growth hack on Twitter or something. I think that there is a risk there that this gets seen as spam in our systems”.