Collaborative filtering

Collaborative filtering is a family of methods for predicting what a person will like by using the recorded tastes of many other people. It is one of the two major techniques behind recommender systems, the other being content-based filtering. The central assumption is that if two people agreed on some items in the past, they are more likely to agree on others than a randomly paired stranger would be. A typical system asks users to rate items such as books, films or recordings, finds the users whose rating patterns are closest to the active user, and recommends items those similar users rated highly but the active user has not yet rated. Approaches divide into memory-based methods, which compute similarity between users or items directly from the rating data, and model-based methods, which learn a model to predict unseen ratings, often with dimensionality reduction. Many deployed systems combine both. The same idea is also used in a broader sense for filtering sensing, financial and web data through collaboration among many sources.