Facebook News Feed: Choosing the Objective and Signals That Order Each User's Posts

You are a product manager on Facebook's News Feed. When someone opens the app, far more posts are eligible to appear than they will ever scroll through: updates from friends, posts from Pages and groups they follow, and possibly recommended or sponsored content. How would you decide the order in which those posts are shown to a given user? Cover what the ordering should optimize for, which signals you would rely on, how you would turn those signals into one ranked list, and how you would tell whether a new ranking is better than the current one. ```hint Objective before signals A list of signals is not yet a ranking. Decide first what a successful feed visit means for the person scrolling, and for Meta, because that choice determines how much each signal should count. ``` ### Clarifying Questions - Are we ranking for an established user with years of history, or also for someone who just joined and has few friends and little activity? - Are ads in scope, or does a separate system choose the ads while we order only the organic posts around them? - Are we designing the ranking approach from first principles, or changing today's ranking to move a specific goal, such as more conversation between friends rather than more time spent? - Which content counts as a post here: friend updates, Page and group posts, videos and links, and recommendations from accounts the user does not follow? ### What a Strong Answer Covers - A stated objective for the ranking that weighs the user's long-term value against short-term engagement, with the reasoning for the choice. - Organized ranking signals: closeness to the author, the user's preferences by content type and topic, the post's own quality and early response, and freshness, with how each would be measured. - A clear way to combine signals into one score, plus rules applied on top of it such as variety, integrity demotions and where ads may appear. - How the approach changes for distinct segments, notably users with little history. - An evaluation plan built on controlled experiments, with a primary metric, guardrails and a horizon long enough to see retention effects. ### Follow-up Questions - Ranking by predicted engagement tends to lift sensational or divisive posts. How would you detect that and correct for it? - An experiment raises time spent but lowers the number of comments friends leave on each other's posts. Would you ship it, and why? - How would you order the feed for someone who joined yesterday and has only a handful of friends?