Propose an ads recommendation model for shop ads
Company: Meta
Role: Data Scientist
Category: Machine Learning
Difficulty: medium
Interview Round: Technical Screen
You need to propose a modeling approach for **recommending/ranking shop ads** (i.e., which shop ads to show and in what order) for a marketplace app.
Describe an end-to-end ML solution:
- What is the **prediction target** (CTR, CVR, expected revenue, expected GMV, multi-objective)?
- What are the **training labels** and how do you handle **position bias** / feedback loops in logged data?
- What features would you use (user, shop, query/context, geo, time) and how would you handle cold start?
- What model family would you start with (GBDT, deep model, two-tower retrieval + ranker, contextual bandit)? Why?
- How would you evaluate offline (metrics, validation scheme) and online (A/B), and what guardrails would you add?
- Name at least 3 failure modes (bias, exploitation vs exploration, latency, calibration) and mitigations.
Quick Answer: The prompt evaluates expertise in recommender and ranking systems—covering prediction target selection (CTR/CVR/expected revenue), label construction and position-bias correction, feature engineering across user/shop/query/context, cold-start strategies, model-family trade-offs, and offline/online evaluation with guardrails; category/domain: Machine Learning for Data Scientist. It is commonly asked because marketplaces require end-to-end, production-ready ad-ranking solutions that balance revenue, relevance, and exploration–exploitation tradeoffs; abstraction level: system-level applied ML and production design.