Social Feed Ranking: Problem Framing & Requirements

If you have worked through a video recommendation ranking system, juggling multi-objective formulation, re-ranking layers, and fairness constraints, social feed ranking will feel familiar in its bones but foreign in its skin. Social feeds share the core ranking principles, yet they introduce a distinct set of challenges that make this one of the most frequently asked ML system design questions across every MAANG company. The social graph injects signals that don’t exist in pure content recommendation. Real-time freshness demands are measured in hours, not days. And the content itself is wildly heterogeneous, spanning text posts, images, reshares, polls, and interleaved ads, all competing for the same viewport.

Consider the interview prompt that launches this case study: “Design the feed ranking system for a platform like LinkedIn or Facebook.” This single question tests product thinking, ML modeling intuition, scale reasoning, and trade-off articulation simultaneously. That density is exactly why interviewers love it.

This lesson focuses exclusively on problem framing and requirements gathering. We will define the business metrics that anchor every downstream decision, scope the scale and freshness constraints unique to social feeds, and compare how candidates at different seniority levels should approach this opening phase. Data strategy and feature engineering come in the next lesson.

Framing feed ranking as a business problem

The first move in any ML system design interview is translating a vague product ask into measurable business objectives. Jumping straight into model architecture without grounding it in business goals is one of the fastest ways to lose points. For social feed ranking, three core business metrics define the optimization landscape.

Defining the three core metrics

Each metric captures a different dimension of platform health, and understanding their interactions is more important than memorizing their definitions.

• Engagement refers to session time, clicks, comments, and reshares that drive ad revenue and platform stickiness. It is the most directly measurable metric and the one most ranking systems default to optimizing. The proxy ML objective is typically $P(\te$ ...