Course contentsShow
← Machine Learning and Deep Learning
Lesson 2349 of 3,53851. Recommender SystemsFree lesson

Collaborative Filtering Overview

Introduction to collaborative filtering: leveraging user-item interaction patterns to predict preferences.

Collaborative Filtering Overview

What you'll learn: How collaborative filtering uses patterns in user-item interactions—rather than item features—to predict what you'll like based on what similar users enjoyed.

The Core Idea

Unlike content-based filtering (which you just learned), collaborative filtering doesn't care what an item is—it only cares about who interacted with it. The fundamental insight: if you and I have rated many movies similarly in the past, we probably have similar tastes, so movies you loved (that I haven't seen) are good recommendations for me.

Think of it like this: imagine you're at a party and someone says, "Oh, you loved Inception and Interstellar? So did my friend Alex! Alex also raved about Arrival—you should watch it." You're relying purely on overlapping preferences, not on whether these movies are sci-fi or have similar plots.

How It Works

Collaborative filtering builds a user-item interaction matrix—typically ratings, clicks, or purchases. It then identifies patterns:

  • User-based: Find users similar to you, recommend items they liked
  • Item-based: Find items similar to ones you liked, based on who else liked them

The power lies in discovering hidden connections: maybe users who like both thriller novels and cooking shows tend to enjoy true-crime podcasts—a pattern no content features would reveal.

Why It Matters

Collaborative filtering excels at serendipitous discovery—surfacing items you'd never find through content features alone. It's the engine behind "Users who bought this also bought..." and "Because you watched..." recommendations.

However, it struggles with the cold start problem: new users with no history and new items with no interactions can't be recommended effectively yet.

Key Takeaway: Collaborative filtering predicts preferences by finding patterns in user-item interactions, leveraging the "wisdom of the crowd" rather than item characteristics—enabling surprising, behavior-driven recommendations.