Try a free sample class
How does an app decide what you see next? In about ten minutes you will rate five movies, work out a recommendation the way a computer does, and check what you learned. No sign-up.
Representative video.
- About 10 minutes
- Works on your phone
- No sign-up. Your answers stay on your device
- Written for class 12 students
What a recommender does
Open a video app, a shopping app or a music app, and much of what you see was picked for you. A recommendation system made that pick: its best guess at what you will watch, buy or play next.
Video apps line up the next video before this one ends.
Shopping apps show products that people bought together.
Music apps build playlists around what you already play.
Two ways to make the guess
You liked a cricket drama, so here is another cricket drama.
Engineers call this content-based filtering.People who rated movies the way you did loved this one, so you probably will too.
Engineers call this collaborative filtering.Rate five movies
How much would you enjoy each one? Tap the stars. Rate at least 3. The movies are made up for this lesson, so go by the kind of film.
Your own recommendation is worked out on this page with JavaScript, which is switched off. Step 3 shows a worked example instead.
0 of 5 rated
How the computer decides
It compares you with four other viewers, then listens to the two whose taste is closest to yours. Three small steps:
- Compare
For each viewer, take the gap in stars on every movie you rated, and average the gaps. A gap of 0 means you agree; a gap of 4 is as far apart as 1 star and 5 stars.
Match = 1 − (average gap ÷ 4)
- Predict
Keep the two viewers who match you best; engineers call them your nearest neighbours. For each movie you have not rated, multiply their stars by their match, add them up, and divide by the total of the two matches.
Prediction = (match × stars + match × stars) ÷ (match + match)
- Recommend
Suggest the movie with the highest predicted stars.
Below: an example student who loves cricket dramas and college comedies.
| Movie | Example | Viewer A | Viewer B | Viewer C | Viewer D |
|---|---|---|---|---|---|
| Last Ball SixCricket drama | 5 | 5 | 2 | 1 | 2 |
| Canteen ChroniclesCollege comedy | 5 | 4 | 5 | 2 | 1 |
| Beyond SaturnSpace adventure | 1 | 2 | 1 | 5 | 3 |
| The Locked HaveliHorror film | 1 | 1 | 2 | 3 | 5 |
| Letters in the RainLove story | 3 | 3 | 4 | 2 | 1 |
| Not rated yet: the computer predicts these | |||||
| Six Balls to GoCricket drama | 3.6 | 5 | 2 | 1 | 3 |
| Backbench HeroesCollege comedy | 4.5 | 4 | 5 | 1 | 1 |
| Signal from TitanSpace thriller | 1.5 | 2 | 1 | 5 | 3 |
| The Third KnockHorror film | 1.5 | 1 | 2 | 3 | 5 |
| Match | – | 90% | 75% | 30% | 25% |
Stars from 1 to 5: the darker the green, the more stars. Dark headings mark your two closest viewers; only their stars go into the predictions (dashed boxes).
Backbench HeroesA college comedy
Predicted 4.5 stars from the two closest matches: Viewer A (90%) gave it 4 stars and Viewer B (75%) gave it 5 stars.
(0.90 × 4 + 0.75 × 5) ÷ (0.90 + 0.75) = 4.5What real apps add
Real apps use the same idea you just worked through by hand, with a few things added.
Comparing everyone with everyone is far too slow, so real systems use shortcuts to find close matches fast.
What you watch to the end, skip, search for or come back to can all count as signals.
Machine-learning models find patterns in taste from the data, instead of a formula written by hand.
Teams often mix in new and different picks, so people do not see only one kind of thing.
Three quick questions
Pick an answer to see if you are right. Nothing is marked or saved.
Show the answer
Their tastes match, so what one loves is a good guess for the other. A close match on the movies they both rated is the whole idea. Each one’s other favourites become good guesses for the other.
Show the answer
Viewer A. The closer the match, the more a viewer’s stars count. Only the closest viewers are used at all, and each one’s stars are multiplied by their match.
Show the answer
Not yet: it needs some ratings first. This is called the cold-start problem. Real apps also use what a movie is about, such as its genre and language, until ratings come in.
From ten minutes to a real recommender
In the B.Tech in CS and AI, the semester 4 module “Machine learning and data preparation” includes a project: a movie recommender trained on real ratings. The same idea as today, with real ratings instead of made-up ones.
Machine learning and data preparation
- Supervised and unsupervised learning
- Regression and clustering
- Accuracy, precision and recall
- Cleaning data and building features
Project: A movie recommender trained on real ratings
Backends and APIs, Data, Machine-learning basics
- The ordering backend of a food-delivery style app
- A price tracker that alerts you when prices drop
By the end of the year: You can build the back end of a real app and train a simple model.
The plan for 2027. The detailed curriculum is published with the partner campus.