A recommendation system predicts which products, videos, articles or actions may be useful to a person. It can use item features, behavior from similar users, the current session and an explicit query to score and rank available options.
Recommendations influence what people see, so evaluation should measure more than clicks. Relevance, diversity, novelty, fairness and long-term satisfaction matter. A personalized system also needs limits on data collection and a way to avoid reinforcing one mistaken assumption.
ELI5
A recommendation system helps narrow a large set of choices to the options that seem most useful for one situation. It combines what is known about the items with clues about the user's needs.
For example, a film service may recommend a short comedy after noticing that a viewer wants something light and has only half an hour. A good recommendation explains the fit instead of merely showing a popular title.
How does a recommendation system work?
It represents users, items or session context and predicts a relevance score for candidate items. The highest-scoring eligible options are then ranked or displayed.
What is the difference between search and recommendations?
Search responds primarily to an explicit query. Recommendations can use broader context and behavior to surface items even when the user has not named them directly.
