Recommender Systems
How does the app know what you will like? Mostly not by understanding movies — by understanding overlap. Collaborative filtering finds users whose past ratings echo yours and borrows their opinions of things you have not seen: people like you also liked this. No genre labels, no plot analysis, just the pattern in the ratings matrix. Its famous weakness is the cold start — a brand-new user or item has no overlap to work from — which is why content-based methods (recommend by item attributes) exist, and production systems blend both plus learned embeddings of users and items.
Rate a few movies. The system scores every other user by how closely their ratings match yours, then predicts your missing ratings as a similarity-weighted average of theirs — watch the recommendation flip as your taste profile changes.
Rate these three movies (tap stars):
Rate at least two movies to see recommendations appear.
No genre labels anywhere: the system never learns what a heist movie is. It only finds users whose past ratings echo yours and borrows their opinions of what you have not seen — collaborative filtering.
Check yourself
Give the first movie five stars and the second one star, then reverse them. Different users become your neighbours. What is the system actually matching on?
Go deeper (free): Google — Recommendation systems course ↗