How Recommendation Systems Work

Two classic approaches

• Content-based — recommend items similar to what you liked, using item features (or embeddings).
• Collaborative filtering — recommend what similar users liked, learning from interaction patterns alone.

The modern twist

Both increasingly rely on embeddings: represent users and items as vectors so "similar" becomes a nearest-neighbor lookup — the same machinery behind semantic search.

Watch out for

The cold-start problem (new users/items with no history) and feedback loops that keep recommending the popular. Start simple, measure with held-out interactions, and add complexity only when it pays off.