Cold Start Problem
When a recommender has too little data to predict well, either for a new user or for a new item with no ratings. It is a chicken-and-egg problem: you need ratings to recommend well, but need good recommendations to get ratings.
When to use
Content-Based Filtering copes with new items because item features exist from day one, while Collaborative Filtering struggles because it needs rating history. The usual fix is a Hybrid Recommender (the Seret+ plan: most popular first, content-based after 3 ratings, blend in collaborative filtering after 10).
See Lec 6.