You are building a recommendation feed for a consumer app. The system works well for active users and established content, but new users and newly created items have little or no interaction history, so ranking quality drops early in their lifecycle.
How would you handle the cold start problem in a recommendation or ranking system?
Cold start for both users and itemsRetrieval, ranking, and re-ranking designFeature store and online versus batch featuresExperimentation and cohort-based evaluation