Design a News Feed Ranking System

Problem Formulation#

Clarifying the ML Objective#

ML framing: Given a user and a pool of candidate posts, predict a composite engagement score for each post so the feed can be ranked by decreasing score.

The business goal is keeping users engaged with content they find valuable. The ML task is pointwise scoring: for each (user, post) pair, predict the probability of each engagement action (click, like, comment, share, long dwell), then combine those predictions into a single ranking score. This is fundamentally a multi-label classification problem dressed up as a ranking problem.