Collaborative filtering
Recommender approach that uses patterns across many users and items (who rated what) rather than explicit content features alone.
Collaborative filtering recommends items by learning from collective behavior: if users with similar histories liked the same movies, the system suggests new titles they have not seen. Early Netflix algorithms such as Cinematch used neighborhood models; prize-winning teams advanced matrix factorization and neural models such as RBMs (Netflix Prize article).
The method is a form of supervised learning on sparse user-item ratings and powers many recommender systems in streaming and e-commerce.