What is ranking vs retrieval in recommendation systems?
Updated May 16, 2026
Short answer
Retrieval selects candidate items, ranking orders them by relevance.
Deep explanation
Recommendation pipelines typically have two stages: retrieval (fast filtering of millions of items) and ranking (precise scoring of few hundred items). Retrieval uses approximate methods, while ranking uses complex ML models.
Real-world example
YouTube first retrieves videos, then ranks them on homepage.
Common mistakes
- Using heavy models for retrieval stage.
Follow-up questions
- Why two-stage architecture?
- What is candidate generation?