User-Based Collaborative Filtering (Dummy Data)
Building a recommender system that suggests items based on similar users’ preferences - the foundation of ‘users who liked this also liked’ features.
User-based collaborative filtering powers recommendations like “users who liked this also liked…” on Netflix or Amazon. Today, I built a simple version using dummy movie data after finishing content-based filtering earlier.
Core Concept
- This method assumes similar users share tastes
- Finds “neighbors” for a target user
- Suggests items those neighbors rated highly but the target hasn’t seen
Implementation Steps
Gathered dummy user-movie ratings data
Created a user-item matrix (rows=users, columns=movies, cells=ratings 1-5)
Computed user-user similarity matrix using cosine similarity
Identified top similar users (neighbors) for the target user
Recommended unseen movies from neighbors’ high-rated lists, weighted by similarity
Check my code: simple-recommender.ipynb
Quick Example
| User | Movie1 | Movie2 | Movie3 |
|---|---|---|---|
| A | 5 | ? | 4 |
| B | 4 | 5 | 4 |
| C | 1 | 2 | 1 |
User A is similar to B (both love Movie1/3), so recommend Movie2 (B’s 5-star pick).
Why It Excites Me
- Bridges to real-world systems like Netflix/Amazon
- Next up: matrix factorization, handling sparsity
- Hands-on dummy data clarified “wisdom of the crowd” magic!
Feeling pumped to keep exploring recommender systems!