Recommendation Engine — WhizzStep AI Lab

How Collaborative Filtering Works

⭐

Collect Ratings

Users rate items (1–5 stars). These ratings form a giant matrix — rows are users, columns are items.

👥

Find Similar Users

Compare your ratings to everyone else's using cosine similarity. High similarity = shared taste.

🔍

Fill the Gaps

Your "taste twins" rated items you haven't seen. Predict your rating based on their scores.

🎯

Recommend Top Items

Show the highest predicted-rating items you haven't rated yet. That's your personalised recommendation!

🎬

Wizzy the AI Tutor

Welcome to the Recommendation Lab! 🎉 Rate at least 6 movies by clicking the stars. Your ratings will be compared to other users to find your "taste twins" — people who like exactly what you like. Rate honestly — the recommendations depend on it!

Step 1 — Rate These Movies

0 / 12 rated

⭐ Your Ratings

Rated0

Avg rating—

Highest rated—

Lowest rated—

💡 Rating guide:
⭐ = Hated it
⭐⭐ = Didn't like
⭐⭐⭐ = It was OK
⭐⭐⭐⭐ = Liked it
⭐⭐⭐⭐⭐ = Loved it!

Leave items blank if you haven't seen them.

🎬

Wizzy the AI Tutor

Look at your taste twins! 👥 We compare your ratings to 8 other users using cosine similarity — a maths formula that measures how aligned two rating vectors are. A score of 1.0 = perfect match, 0 = completely different taste!

Step 2 — User Similarity Scores

📊 Similarity Stats

Most similar—

Similarity score—

Least similar—

Your ratings used—

Cosine Similarity:
Treats each user's ratings as a vector in N-dimensional space. If two vectors point in the same direction, their cosine similarity is close to 1!

🎬

Wizzy the AI Tutor

Here are your personalised recommendations! 🎯 These are movies your taste twins loved but you haven't rated yet. The predicted score is a weighted average of their ratings — users who are more similar to you have more influence. Does this match what you'd actually enjoy?

Step 3 — Your Personalised Recommendations

🎬

Wizzy the AI Tutor

This is the rating matrix — the data structure behind every recommender system! Rows are users, columns are movies. Empty cells are what the algorithm tries to predict. Notice how sparse it is — most users have only rated a few items. That's called the sparsity problem!

Step 4 — The Rating Matrix

⬜ Not rated ⭐ 1 ⭐⭐ 2 ⭐⭐⭐ 3 ⭐⭐⭐⭐ 4 ⭐⭐⭐⭐⭐ 5

🎬

Wizzy the AI Tutor

🎊 You've just built the algorithm behind Netflix, Spotify, Amazon, and YouTube! Collaborative filtering powers billions of recommendations every day. You understand user-based CF, cosine similarity, and the sparsity problem. That's real ML engineering!

🎬

Recommender Badge!

You built a collaborative filtering recommendation engine from scratch!

🎬 WhizzStep AI Lab

This certifies that

Student Name

has built a Collaborative Filtering Recommendation Engine

Recommender Expert

Cosine Similarity

Netflix Engineer

whizzstep.in

Key Concepts Mastered

Collaborative Filtering

👥 Taste Twins

"Users who liked what you liked also liked X." No content analysis needed — just patterns in ratings.

Cosine Similarity

📐 Vector Alignment

Treats ratings as vectors. cos(θ) between two vectors = similarity. 1.0 = same direction = same taste.

Sparsity Problem

🕳️ Missing Data

Most users rate very few items — the matrix is 99% empty. Matrix factorisation (SVD) solves this.

Cold Start

🥶 New User Problem

A brand new user has no ratings — can't find taste twins. Fix: ask for preferences at sign-up.

Filter Bubble

🫧 Echo Chamber

Recommenders only show what you already like, narrowing your experience. A real ethical concern!

Content-Based

🎭 The Other Approach

Instead of user similarity, analyse item features (genre, director). Netflix uses both methods together.