Beyond newest first
A reverse chronological feed is simple but often shows low value posts just because they are recent. A ranked feed scores each candidate post and orders by score, so the most engaging content rises.
Ranking takes a set of signals as input and combines them into one number per post.
Common signals
- Recency. How fresh the post is. Older posts decay.
- Affinity. How close the viewer is to the author, based on past interaction.
- Engagement. Likes, comments, and shares the post is gathering.
- Content type. Whether the viewer tends to engage with video, photos, or text.
- Negative signals. Hides, reports, and quick scroll past lower a post.
How signals combine
Early systems used a fixed weighted formula. Modern systems feed the signals into a machine learning model that predicts the chance the viewer will engage, then ranks by that predicted value.
The output is a ranked candidate list, which later stages may further filter for diversity and policy.
Key idea
A ranked feed scores candidates from signals like recency, affinity, and engagement, often via a model that predicts viewer interaction, instead of pure recency.