An introduction to recommendation systems and how they help us choose from millions of online options. These systems act like digital assistants that learn what we enjoy.

Meet Your Personal Guide

Known as explicit feedback, this is when you directly tell a system what you think using stars or thumbs up. It is the clearest way for a computer to learn your preferences.

Direct Ratings and Reviews

Known as implicit feedback, this is when the system learns from what you do, like clicking a link or finishing a video. It allows the computer to understand your taste without you having to say a word.

Learning from Your Actions

Content-based filtering suggests items similar to ones you liked before by looking at descriptions like genre, color, or author. If you liked one mystery novel, the system will look for more books in that same category.

Matching Based on Features

Collaborative filtering makes suggestions by looking at the behavior of many other people. It assumes that if you and another person liked the same five movies, you will probably like the sixth one they watched.

Wisdom of the Crowd

User-based filtering finds other people with very similar habits to yours. By seeing what your 'taste twins' are enjoying, the system can predict what you might want to try next.

Finding Your Taste Twins

Item-based filtering looks for products that are often bought or liked by the same groups of people. This is why you often see suggestions like 'frequently bought together' when shopping online.

Items That Go Together

The 'cold start' problem happens when a system has a brand-new user or a new product with no history. Without past data to look at, the system has to find clever ways to make its first guess.

The Empty Profile Problem

Hybrid systems combine multiple techniques, like looking at both item descriptions and other users' habits, to get the best of both worlds. Most modern apps use this multi-layered approach.

Mixing Different Methods

Computers can find 'latent features,' which are subtle connections between items that humans might not notice. For example, it might realize you like a specific visual style that doesn't have a formal name.

Discovering Hidden Patterns

Context-aware systems adjust their suggestions based on where you are or what time it is. You might get upbeat workout music in the morning and relaxing beats in the evening.

Right Time and Place

Serendipity is when a system suggests something unexpected that you wouldn't have searched for but end up loving. It prevents your recommendations from becoming boring or repetitive.

The Joy of Discovery

Companies use math to see if their recommendations are accurate or if people are just ignoring them. They track how often a suggestion is clicked to improve the system over time.

How We Measure Success

Sometimes recommendation systems can keep us trapped in a loop of the same ideas, known as a filter bubble. It is important to design systems that encourage variety and fairness.

Why Measure Success?