TikTok explains its For You feed
On 18 June 2020 TikTok published an account of how its recommender system fills the For You feed: it ranks videos by the user's interactions, video information (captions, sounds, hashtags) and, with lower weight, device and account settings. By the post, an account's follower count and its previous high-performing videos are not direct factors.
Why it matters
A platform whose feed consists almost wholly of recommendations named the groups of signals, and that watching a longer video to the end weighs more than weak matches such as a shared country. The post also acknowledges the risk of a 'filter bubble' and describes how the system diversifies the feed. It contains no figures.
The post says the feed generally does not show two videos in a row with the same sound or creator, does not recommend duplicates or what was already seen, and asks new users to pick categories of interest or shows them popular videos. It ends by saying that at the Transparency Center in Los Angeles invited experts will be able to learn how the algorithm works and review source code; it gives no opening date. The post has no figure, of scale or of effect, and no document was found in which anyone independent confirmed what it describes. The record states the disclosure, not its accuracy.