In Depth
Collaborative filtering is the most widely used approach in recommendation systems. It comes in two main forms: user-based (finding users with similar preferences and recommending what they liked) and item-based (finding items similar to those a user has already liked based on co-occurrence patterns). Matrix factorization methods like SVD decompose the user-item interaction matrix to discover latent factors.
The key advantage of collaborative filtering is that it requires no knowledge about item features; it works purely from behavioral data like ratings, purchases, or clicks. This makes it applicable across domains without domain-specific feature engineering. Netflix's famous recommendation system and Amazon's 'customers who bought this also bought' are classic collaborative filtering applications.
Challenges include the cold-start problem (new users or items have no behavioral data), data sparsity (most users interact with a tiny fraction of available items), and popularity bias. Modern systems address these by combining collaborative filtering with content-based methods (hybrid approaches) and using deep learning models that can incorporate both behavioral and feature data.