Picking Ranking Signals for Meta's Feed: Viewer, Relationship, Content and Context

Meta's News Feed orders the posts each person sees. What signals would you rely on to make that order more relevant to each person? ```hint Sort signals by what they describe Signals can describe the viewer, the author, the viewer's tie to the author, the post itself, or the moment of viewing. Grouping ideas this way exposes gaps; then ask which signals reflect what a person values rather than what merely catches their eye. ``` ### Clarifying Questions - Does "signals" mean the inputs the ranking model uses to order posts, the metrics that show whether relevance improved, or both? - Which posts are in scope: friends and family, Groups and Pages the person follows, recommended posts and Reels from accounts they do not follow, and ads? - What should relevant mean here: posts people engage with, posts they say they value, or more meaningful conversations with people they know? - Are there limits on data use, such as activity from outside Meta's apps or sensitive personal attributes? ### What a Strong Answer Covers - A working definition of relevance for the feed that goes beyond engagement, tied to Meta's goals. - Signals organized into clear families (viewer, relationship, author, content, context, explicit feedback), each with what it predicts. - A prioritized choice of which signals to add or reweight first, and how predictions combine into a ranking. - A measurement plan with a primary metric and guardrails against bait, misinformation and privacy harm. - How noisy or gameable signals such as likes and time on screen are handled. ### Follow-up Questions - Time spent on a post rises when a post upsets someone as well as when it is valuable. How would you tell the two apart? - How would the signal mix change for a new user with few friends and no history? - Pages start writing posts that bait comments. Which of your signals are exposed, and how would you respond?