I started rattling off engagement signals like likes and comments and shares, and then realized pretty quickly that felt too surface-level.
Start by clarifying the goal of improving relevance—likely increasing meaningful interactions and long-term user satisfaction. Then propose a mix of explicit and implicit signals, prioritizing those that predict long-term value over short-term clicks. Finally, discuss how to validate and iterate on these signals using A/B tests and guardrail metrics.
Pro tip: Emphasize that signals should be evaluated by their causal impact on long-term user retention, not just correlation with engagement. Mention that Meta often uses downstream metrics like 'meaningful social interactions' (MSI) and 'time well spent' to avoid optimizing for clickbait.
Define what 'relevance' means for News Feed: likely showing content that users find valuable and engaging over time. Align on north-star metrics like meaningful interactions, retention, and user satisfaction.
Break signals into explicit (e.g., likes, comments, shares, hides, reports) and implicit (e.g., dwell time, scroll depth, video watch time, click-through rate). Also consider content-based signals (e.g., topic, recency, source authority).
Evaluate which signals best predict long-term user value. For example, comments and shares may indicate stronger interest than likes; dwell time can signal quality even without clicks. Downweight signals that encourage clickbait or passive consumption.
Propose A/B tests to measure the impact of new signals on key metrics. Include guardrail metrics (e.g., user reports, hide rate, survey satisfaction) to catch negative side effects.
Discuss potential trade-offs: short-term engagement vs. long-term satisfaction, diversity of content vs. relevance, and fairness across creators. Mention how to handle cold-start users or sparse data.
AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.