A feed can feel uncannily personal when several items in a row match an interest. It can also feel strangely insistent, returning to a subject after curiosity has passed. Both experiences make more sense when recommendation is understood as a process of prediction.
TikTok’s explanation of its recommendation system describes using signals such as interactions and information about content to rank videos. This is the platform’s account of its own system, not an independent audit of every decision it produces. [1]
A signal needs interpretation
Watching something can mean enjoyment, confusion, disagreement, or simple inability to look away. A system may record the behavior more readily than the reason behind it. The prediction can be useful without accurately describing the viewer’s values.
Repeated recommendations also change the environment in which later signals arise. If the next screen offers more of a subject, the viewer has more opportunities to interact with it. What looks like a stable preference may partly reflect the choices that were repeatedly made available.
Personalization is an editorial experience
A feed determines what becomes easy to encounter and what remains out of sight. That makes its design culturally significant even when no individual editor chooses every item. The objective used to rank material affects the public experience of the product.
For readers, the practical distinction is between relevance and reliability. A post can be well matched to an interest and still be false, incomplete, or unrepresentative. A personalized sequence also cannot establish how common a belief is outside that sequence. The feed is a selection made through a system with particular aims. Remembering that fact creates room to seek another source, change the subject, or step outside the pattern the next recommendation expects us to follow.