Understanding the Netflix Recommendation System

Ever wondered how suddenly you stumble upon a great movie/show on Netflix just because it shows up on your home page? According to Netflix, 80% of TV shows people watch are discovered via Netflix recommendation system. Fascinating, isn’t it? Experts suggest it’s the AI and machine learning at play.

But, how do these algorithms work?

How does Netflix ensure that approximately 167 million of its paid users get unique content from its massive library of 15400 titles?

We will try to address these questions in this read. Hold tight.

What is a Recommender System?

A recommender system analyzes patterns based on consumption habits, preferences, likes-dislikes, and various other parameters to arrive at a set of recommendations for a user to consume. For example, it can predict the rating you would assign to a title/show based on your historical rating data. Of course, the actual recommender systems use sophisticated data analysis and machine learning algorithms to arrive at the suggestions. Nearly all OTT platforms use some form of recommendation system, but what makes Netflix standout is the amount of data it has at its disposal (230 million active users) and the number of titles in its library.

So, how does the Netflix Recommendation System Work?

There are a variety of algorithms that collectively define the Netflix experience, most of which you will find on the home page. Per Netflix, they only have a window of 60 to 90 secs [2] to suggest shows/titles, before a user losses their interest. So, how does Netflix achieve this? we will first try to understand the generic idea of Netflix recommender algorithm and then use it to explain two specific applications namely-

  1. Personalized Video Ranker (PVR)
  2. Because You Watched

Let’s simply understand how this algorithm works. Imagine a tricycle with the first wheel of the algorithm being the user data that Netflix collects. It comprises of show/title watching habits of the user – past/current shows, the time duration, time of day, etc. The second wheel of this algorithm is the data Netflix gathers from its inhouse staff, freelancers who explicitly watch every minute of the show/title and tag it. Sounds like a fancy job? Anyway, these tags could be genre tags ex. crime thriller, type of cast, theme of the show, etc. The tags and the user behavioral data are then combined to derive the most critical part – weighs (or points), the third wheel.

These machine learning algorithms help to answer questions like- How much score should we assign if the title was watched yesterday? Should it weigh (scored) more than the title watched a year back? What weights (scores) to assign if the user abandons the title after ten minutes or binges over it for the entire night. Based on the actions performed by the user, the algorithm scores and assigns weights for each outcome. The higher the score, the likelier for it to show up on your recommendations. All of these wheels need to run in tandem for the tricycle to move forward.

Related: Take a look at the TikTok Algorithm
Personalized Video Ranker (PVR)

A typical Netflix homepage consists of around 35-40 rows and up to 70 odd titles in each row. The numbers, though, change according to the device being used. Videos under each title generally come from the same algorithm. The personalized video ranker (PVR) algorithm drives genre rows such as Crime and Thrillers, TV Action, and Adventure (image below). As indicated by its name, this algorithm orders the entire catalog of videos for each user profile in a personalized manner.

Netflix Homepage
Because You Watched (BYW)

This category of videos anchors its recommendations to a single title/video watched by the user. Generally called as the “sims” algorithm, an un-personalized algorithm that features recommendations based on user’s consumption of a specific show/title. Even though the sims ranking is not personalized, the choice by which BYW rows make it into a homepage is tailored to the content, user previously watched. Because its called ‘Because You Watched’.


Netflix has been using these algorithms to drive its business since the early 2000s when they were in the movie renting space. But it was only after they transitioned into a content platform, that they utilized the true potential of their algorithms. According to Netflix, its personalized recommendation engine is worth $ 1 billion, or in other words, Netflix believes it could lose $1 billion or more every year from subscribers quitting its services if it weren’t for its recommendation engine.

Now that you know more about the recommendation system, we’d recommend you head here and join the discussion. While you’re there also drop a thank you to the author Prasad Antapurkar for this piece.

Join the community of 5000+ folks by subscribing to our newsletter.

One thought on “Understanding the Netflix Recommendation System

Leave a Reply

Your email address will not be published. Required fields are marked *