How Netflix Knows What You Want?

It’s Friday night. You open Netflix. You don’t search. You don’t browse much. Yet somehow… the perfect movie is already there. How?

This is the power of Recommender Systems. The Core Idea is that a recommender system tries to answer one simple question:

“What should we show this user next?”

It learns from:

• What you watched

• What others watched

• Patterns across millions of users

What are Recommender Systems?

Recommender systems are algorithms that predict what a user might like based on data; what they’ve watched, bought, rated, or clicked.

Recommender systems are design to suggest relevant items (e.g., movies, products, songs) to users based on their preferences and behavior.

Netflix, Spotify, and Amazon all use them. A recommender system predicts what a user might like, based on available data. You encounter them daily:

• Netflix → “Because you watched Inception…”

• Amazon → “Customers who bought this also bought…”

• Spotify → “Your Daily Mix”

The core challenge is the “cold start” problem: what do you recommend when you know nothing about a new user or a new item?

Types of Recommender System

There are two main families of techniques:

• content-based filtering (focus on item attributes) and

• collaborative filtering (focus on user behavior patterns).

Most production systems combine both.

Figure: Overview of a Recommender System. Source: media.geeksforgeeks.org.

Approach	Key idea	Data needed	Key Comparison
1. Content-Based	Recommend items similar to what you liked before	Item features (genre, cast, etc.)	User profile vs Item
2. User-Based CF	Find users like you, recommend what they liked	Rating history of all users	User vs User
3. Item-Based CF	Find items similar to ones you rated highly	Rating history of all users	Item vs Item

** CF = Collaborative Filtering