Are recommendation systems enough?

Netflix’s Content Discovery Analysis

Shrusti Ghela
6 min readAug 3, 2022

Last night, I returned home after a tiresome day at work and fixed myself a cup of hot tea. I started browsing Netflix to look for something that I could watch, sit back, and relax.

I finished my tea, and I was still browsing! I didn’t realize that I had spent so much time without actually watching anything. By the end of it, when I decided to watch something, I was so exhausted, that I decided to call it a night and go to bed.

Have you spent more time deciding what to watch than actually watching something?

Choosing a streaming service is another decision in itself!

This question made me think of all those times when I spent hours picking what to watch! And I was horrified. How much time do I spend looking for something to watch? I was curious and decided to look for answers.

Well, the answer will shock you! According to a study from the Ericsson Consumer Lab, “Over the course of an average American’s life, they’ll waste 1.3 years endlessly scrolling through TV guides or browsing a variety of menus in an attempt to find something to watch — and failing.” [1]

If this shocks you and you want to get into the details of this, you don’t have to go anywhere to look for answers! This blog contains all that you seek and some suggestions to turn this problem around!

Let’s refresh our memories with some of the concepts that we will discuss throughout this blog.

Recommendation Systems

“Recommendation system is a subclass of Information filtering Systems that seeks to predict the rating or the preference a user might give to an item. In simple words, it is an algorithm that suggests relevant items to users.” [2]

Netflix, Amazon, YouTube, Spotify, LinkedIn, and many others use such algorithms to provide personalized recommendations to the user as to what to watch, buy, listen to, and who to connect with!

Information Overload

“Information overload describes the excess of information available to a person aiming to complete a task or make a decision. This impedes the decision-making process, resulting in a poor (or even no) decision being made.” [3]

The Paradox of Choice / Choice Overload

“Choice overload is the tendency for people to get overwhelmed when they are presented with a large number of options, often used interchangeably with the term paradox of choice.” [4]

Content discovery

“Content discovery is the process of searching through and finding content. This is a process that users go through when they find and engage with brand’s content.” [5]

Netflix Recommendation Engine (NRE)

Netflix’s motive for using Recommendation Systems was simple — To provide personalized suggestions to users to reduce the amount of time and frustration to find some great content to watch. [6]

So, why do users still spend a lot of time finding something to watch and as a result are frustrated?

The major challenge here is to suggest ways to improve content discovery so that the users get ‘more’ relevant content, based on their preferences. In order to improve content discovery, let’s first look at the actual structure of Netflix’s content discovery platform.

In a quick first look, we detect the difference in the amount of Netflix curated content and the amount of personalized content!

🔴 Netflix curated content 🟢 Personalized content
🔴 Netflix curated content 🟢 Personalized content
🔴 Netflix curated content 🟢 Personalized content

Looking at this, it is only natural to think that things are going in the wrong direction from the user's perspective!

User Survey

Based on this, I started asking other users what they felt about this. And here is what most of them had to say:

  • “Lots of content on display.”
  • “It takes me a long time to decide what to watch.”
  • “I usually watch what my friends recommend.”
  • “I usually watch something that everyone else is watching.”
  • “I end up watching what I have already seen.”

From this short survey, I understood one thing. Most of these users don’t rely on the Netflix Recommendation Engine while deciding what to watch. Most of them watch something that was recommended by someone else or the most viewed show/movie at the time.

Netflix invests millions of dollars to update its algorithm. This is something extremely complex and challenging. But apparently, the users are not using the NRE to select what they watch. At least not yet.

Even though the NRE is the state-of-the-art in the field, it still does not solve the problem of providing personalized suggestions!

Why is this happening?

  • The home page displays a lot of content. This leads to choice overload and the users can’t decide on what to watch.
  • The home page displays a lot of Netflix curated content as compared to personalized suggestions. This causes the user to struggle to find the target content.

How to fix these problems?

  • The users are struggling to find the target content — Make a filter to reduce the amount of irrelevant content!
  • The users watch what their friends recommend — Enable sharing options within Netflix.
Current sharing option vs. Suggested sharing option (only available on the app)
  • The users watch what their friends recommend — Enable the “recommendation” profile section!
Suggested “recommendation” profile section
  • The users watch what their friends recommend — Enable recommending easier by providing a “recommend” button at the end screen!
Suggested “recommend” button
  • The users watch what their friends recommend — add new rows based on what their friends are watching and what they recommend
Suggested additional rows to the home page
  • The users watch what their friends recommend — Enable an “ask for recommendation” option for friends and circles!
  • The users are struggling due to choice overload — Enable 3 “channel” options (similar to the “radio” option in the music apps) that start the shows/movies based on the user preference, recommendation from a friend, or the top shows/movies in the area!
  • The users end up re-watching something that they have watched before — Enable an option to be able to hide already watched content.

One preliminary thing that could be done is to create a link between engineers that develop these recommendation systems and designers who design the platform. Both engineers and designers are doing fantastic jobs for their specific tasks, but there seems to be some connection missing when we look at these tasks under a combined umbrella. Eventually, Netflix needs the final product to work! And working on a link that connects these tasks might just seem to help. (This is essentially what I tried to do with this blog.)

There are some functionalities that are available on the app but not on the browser. It would be great if the browser supports all the functionalities that the app supports! (Most of the users that I interviewed, prefer using the laptop to use Netflix)

These are just some suggestions based on what I experienced, or what answers I received from other users.

I am happy to discuss more!

References:

[1] J. Alexander, Study: Viewers will spend a total of 1.3 years trying to find something to watch (2016), polygon.com

[2] S. Agrawal, Recommendation System -Understanding The Basic Concepts (2021), Analytics Vidhya

[3] Information Overload, Interaction Design Foundation

[4] The Paradox of Choice, The Decision Lab

[5] C. Harding, Content discovery: What it is and how to do it right (2021), GatherContent

[6] Recommendations, Netflix Research

--

--

Shrusti Ghela
Shrusti Ghela

Written by Shrusti Ghela

MSDS @ University of Washington

No responses yet