Skip to content

Recommender Systems

The AI behind personalization — deciding what to show, suggest, or rank for each user. The quiet workhorse of the internet.

A recommender system is the software that decides what to show you next — the next video, song, product, or post. Think of a friend who knows your taste so well they can always pick a song you'll love. Recommenders try to be that friend for millions of people at once.

They mostly work in two ways. The first is "people like you also liked this": if you and someone else enjoyed the same ten things, you'll probably enjoy the eleventh thing they liked. The second is "this is similar to what you already liked": if you loved a spooky mystery movie, here's another spooky mystery. Real apps blend both.

This is why two people open the same app and see completely different things. The system is quietly guessing, learning from every tap and skip, and updating its guess about you all the time.

The main ideas

  • Collaborative filtering — Recommend from what similar users liked — matrix factorization and neighborhood methods.
  • Content-based filtering — Match items to a user from item features and past preferences.
  • Learning to rank — Order candidates by predicted relevance — pointwise, pairwise, and listwise ranking.
  • Two-tower retrieval — Embed users and items separately, then retrieve nearest neighbors at scale.
  • Cold start & the long tail — Recommending for brand-new users and items with little data.
  • Feedback loops & bias — How recommendations shape behavior — popularity bias, filter bubbles, and evaluation pitfalls.

Machine Learning · Information Retrieval & Search · Data & MLOps

Want to make things?

Head to AI School — AI camps where kids build their own games.