Collaborative filtering finds patterns in user-item interactions. "Users like you also liked..."
User-based: Find similar users. Recommend what they liked. Doesn't scale well.
Item-based: Find similar items based on co-occurrence. "Bought together" recommendations. Scales better.
Pros: Discovers unexpected recommendations. No item features needed.
Cons: Cold start for new users and items. Popularity bias. Sparse matrices.
Interview question: "How compute user similarity?"
Cosine similarity on rating vectors. Pearson correlation for centered ratings.