Model an ads ranking system
Company: Snapchat
Role: Machine Learning Engineer
Category: Machine Learning
Difficulty: medium
Interview Round: Onsite
## Scenario
You are designing the **modeling approach** for an ads ranking system in a feed/search product.
## Requirements
- For each ad impression opportunity, choose and rank candidate ads.
- Optimize business value (e.g., revenue) while maintaining user experience.
- Account for auction/bidding constraints (e.g., advertisers bid per click or per conversion).
## What to cover
- What labels you would predict (CTR/CVR/expected value), and how you combine them into a final ranking score.
- Feature sets (user, ad, context) and handling sparse/categorical data.
- Training data generation, delayed feedback, and bias (position bias, selection bias).
- Calibration, evaluation metrics, and online experimentation.
- Cold start for new ads/advertisers and exploration.
Quick Answer: This question evaluates machine learning modeling, feature engineering, and systems-level ranking competencies for ad selection and monetization, covering prediction targets (CTR/CVR/expected value), handling sparse and categorical features, delayed feedback and bias, calibration, evaluation metrics, and online experimentation.