Recommendation system

A recommendation system selects possible content from a large pool and puts it in an order intended to be relevant to a person. It can combine several algorithms with fixed selection rules and trained models. The models might learn from past views or reactions which content is likely to be interesting.

An example: The home page of a video platform. From a huge pool of videos, a first stage pulls a much smaller selection of possible candidates. Another stage ranks that selection so that what you’re most likely to find interesting ends up at the top. Suggestions in music apps and product recommendations in online shops come about in a similar way.

Not to be confused with a single algorithm: When people talk about a platform’s “algorithm,” they often mean this entire system, not one individual algorithm or model. It also differs from a spam filter: the filter judges a single email, while a recommendation system puts many items in order.

Where you’ll come across it: Wherever an app suggests content you didn’t search for: in social media feeds, in “You might also like” rows on streaming services, in online shops. And in debates about which content platforms show to whom, and why.

The underlying terms, algorithm and model, are explained in Program, Algorithm, Model Compared.