seniorRecommendation Systems
What is real-time feature store in recommendation systems?
Updated May 16, 2026
Short answer
A feature store manages and serves features consistently for training and serving.
Deep explanation
Feature stores ensure that offline training and online inference use identical feature definitions. Real-time feature stores update user behavior signals instantly and serve them with low latency for ranking models.
Real-world example
TikTok updating user interest vectors in real time.
Common mistakes
- Training-serving skew due to inconsistent features.
Follow-up questions
- What is training-serving skew?
- Why feature store is needed?