Collaborative Filtering
Collaborative filtering is the recommendation technique that predicts what a user will like from the behavior of many others. We explain its principle, its contrast with content-based filtering, its neighborhood and matrix-factorization approaches—popularized by the Netflix Prize—its known problems and its neural version.
Collaborative filtering is a recommender-system technique that predicts a user's interests from the behavior of many other users. Its principle is intuitive: if two people agreed in their tastes in the past, they will probably agree in the future. It works on a user-item interaction matrix—ratings, clicks, purchases—and tries to fill in its gaps; unlike other methods, it looks not at the content of the items but at behavior.
Versus content-based filtering
It is worth contrasting it with content-based filtering, which recommends items similar to those already liked, according to their attributes (genre, topic, etc.). Collaborative filtering, by contrast, can suggest something outside the user's known profile—what similar others liked—at the cost of needing more behavioral data to work.
Neighborhood and factorization
There are two major approaches. The neighborhood-based (or memory-based) one seeks similarities: the user-based version recommends what similar users liked, and the item-based version, items similar to those you already consumed. The model-based one, and in particular matrix factorization, decomposes the large interaction matrix into low-dimensional latent factors that capture tastes and characteristics. The latter became widely known with the Netflix Prize (2006-2009), a one-million-dollar contest to improve recommendations by 10%, won by the BellKor's Pragmatic Chaos team in 2009.
Its problems
The method carries known difficulties. The cold start: a new user or item, with no history, is hard to recommend for. Sparsity: the matrix is almost empty, because each user interacts with a tiny fraction of the catalog. And scalability, as the number of users and items grows.
The neural version
Recent research has brought collaborative filtering into deep learning. Neural collaborative filtering (He and colleagues, 2017) learns embeddings—vector representations—of users and items and replaces factorization's simple product with neural networks that capture nonlinear relationships. In 2026 recommenders are usually hybrid, combining behavioral and content signals.
This article was produced with artificial intelligence under human editorial oversight.
