Build cold-start restaurant ratings
Company: Uber
Role: Data Scientist
Category: Machine Learning
Difficulty: medium
Interview Round: Technical Screen
Uber Eats wants a cold-start rating system for newly onboarded restaurants before they accumulate enough real reviews. You are asked to design the modeling approach rather than name a specific algorithm.
Describe:
1. What target variable you would predict and how you would construct training labels.
2. What features you would use at restaurant launch time. Consider merchant metadata, cuisine, price tier, location, chain history, delivery operations, photos and menu completeness, similar restaurant signals, and neighborhood demand.
3. How you would avoid leakage and selection bias when training the model.
4. How you would validate model quality, calibration, and fairness across merchant types and geographies.
5. How you would represent uncertainty so the business does not over-trust noisy predictions.
6. How the model would integrate with experimentation to measure whether the predicted ratings actually improve marketplace outcomes.
Quick Answer: This question evaluates a data scientist's ability to design a production-ready predictive modeling approach for cold-start ratings, testing competencies in defining target variables and labels, selecting launch-time features, preventing leakage and selection bias, validating calibration and fairness, representing uncertainty, and integrating models with experimentation. It is commonly asked in the Machine Learning domain to assess system-level thinking about model validity and marketplace impact, operating at a high level of abstraction that blends conceptual modeling design with practical product-integration and evaluation considerations.