• Miranda Soong

Why Your Netflix is Unique to You

April 18th, 2022

By Miranda Soong

You’ve probably noticed that when you open Netflix, your landing page looks entirely different from how your friend’s page looks. When you scroll through all the movies and TV shows, you might see more action and thriller content appearing on your page, while your friend’s might be filled with more romance and period pieces. You are essentially looking at your own special version of Netflix that is specifically curated to you and your preferences. With the abundance of choices that Netflix offers, sometimes it’s difficult to choose something to watch right away. To solve this issue, the company utilizes complex machine learning algorithms and a variety of artwork in the thumbnails to personalize your page to optimize the chances of you clicking to watch the titles it recommends to you. Let’s take a look at exactly how Netflix harnesses data and artwork to strengthen the user experience and recommend the best content to you.


Ranking Algorithms for Personalized Recommendations

We all know that Netflix curates its recommendations based on its learning from our activity on the platform like viewing history and streaming hours. You might have also clicked on a title’s thumbnail and seen that it was a “96% Match” for you––a clear example of Netflix’s algorithms at work to show you titles you will most likely click on. Netflix utilizes a set of powerful ranking algorithms that are all optimized for various purposes, like the “Top Picks” row that show user-specific recommendations and the “Trending Now” and “Top 10 in the U.S. Today” rows that analyze recent popularity trends amongst its subscriber base. In order to construct these algorithms, Netflix utilizes a two-stage online experiment process using an “interleaving” technique and A/B testing metrics.

Courtesy of Netflix Technology Blog


The first stage consists of a quicker “pruning” step in which Netflix generates an intermixed blend of two ranking algorithms to present choices from each algorithm side-by-side. Additionally, Netflix factor in position bias as users are more likely to choose a title closer to the left of the page, so it uses a technique it call “team draft” interleaving to alternate the titles from each ranking algorithm in a fair way. Because interleaving only enables Netflix to quickly identify relative member preferences from its ranking algorithms, Netflix utilizes a second stage of experimentation with A/B testing to measure longer-term user behavior, like member retention. With this experimentation process, Netflix is able to quickly measure its data and metrics to determine the best candidates of content for a user within days; thus, accelerating its rate of machine learning to actively improve its rankings.


Variety in Thumbnail Artwork Design

Netflix’s powerful ranking algorithms are not the only factor in getting a user to click on a title to watch it. You have likely clicked on a title due to the eye-catching visual artwork in the thumbnail. On Netflix, the visual artwork is the top influence for enabling viewers to watch––after all, visuals are literally the central aspect of movies and TV shows. Knowing this, Netflix believes that it only have 90 seconds to capture the attention of its viewers before they move onto another activity. According to Netflix’s consumer research studies, thumbnail artwork constitutes 82% of a user’s focus while browsing Netflix. Furthermore, Netflix recognizes that each user spends as little as 1.8 seconds considering each title. Netflix’s Creative Services team has mastered the practice of effectively combining its ranking algorithms and viewer preferences with visual artwork by using a contextual bandits approach to A/B testing and aesthetic visual analysis to design multiple variants of thumbnail artwork for one title.

Courtesy of Netflix Technology Blog


For instance, let’s use the film Good Will Hunting. If you watch a lot of content within the romance genre, the thumbnail Netflix might show you would contain an image of the leads Matt Damon and Minnie Driver about to kiss. However, if you watch a lot of comedy content, your thumbnail artwork might be an image of comedian Robin Williams instead.

Courtesy of Netflix Technology Blog


Additionally, Netflix incorporates regional trends and global preferences in different territories. A great example of how it appeals to viewers in different regions is with the show Sense8, which contains many international actors and storylines. You can see that the thumbnail Netflix displays to viewers in Germany differs from the one shown in the U.S., as Netflix’s machine learning has recognized that viewers in Germany tend to like more abstract visuals while viewers in the U.S. are more attracted to facial expressions and recognizable characters instead.

Courtesy of Netflix Technology Blog


Due to these experiments and improvements in its machine learning, Netflix has learned a lot about its viewership and consumer behavior and use this information to curate the most attractive and clickable artwork for each individual viewer. Netflix’s trends show that viewers are more likely to click on a title when the artwork contains individual actors over group cast shots, recognizable and polarizing characters, and complex facial expressions.


The Powerful Merging of Creativity and Technology

Netflix’s data-driven approach to product development and user experience is a testament to how marketing is not just about advertising, promoting, and selling a product. Marketing is about understanding your consumers’ needs and developing a product that can bring value to them. Now, it’s also shifting towards becoming a science with the advancement of technology in creating a product that can best satisfy those consumer needs. Netflix does just that. Its ranking algorithms combined with its visual artwork create an optimal user experience when browsing the platform. Imagery is such a powerful tool when it comes to film, television, and entertainment, and by using the power of visuals and data, Netflix hopes to help you find something you want to watch––and hopefully in under 1.8 seconds.


----------


Read more on the Netflix Technology Blog

If you would like to read more about Netflix’s technology, algorithms, and experimentation processes, I highly recommend checking out the Netflix Technology Blog to learn more! Here are the blog posts I referenced throughout this article:


Innovating Faster on Personalization Algorithms at Netflix Using Interleaving,” by Joshua Parks, Juliette Aurisset, Michael Ramm


Artwork Personalization at Netflix,” by Ashok Chandrashekar, Fernando Amat, Justin Basilico and Tony Jebara


Selecting the best artwork for videos through A/B testing,” by Gopal Krishnan and Nick Nelson


----------


About the Author

Miranda Soong is one of the Co-Content Directors this semester. She is a sophomore at NYU, pursuing Economics with a minor in the Business of Entertainment, Media, and Technology.


Sources

https://netflixtechblog.com/interleaving-in-online-experiments-at-netflix-a04ee392ec55 https://netflixtechblog.com/artwork-personalization-c589f074ad76

https://netflixtechblog.com/selecting-the-best-artwork-for-videos-through-a-b-testing-f6155c4595f6

https://about.netflix.com/en/news/the-power-of-a-picture

https://www.youtube.com/watch?v=axCBA3VD5dQ

https://www.youtube.com/watch?v=JwHzqYN0WTA


41 views0 comments

Recent Posts

See All