How Does Netflix Come Up With Recommendations For You? It's Actually A Two-Part Trick
Netflix is kind of like a magician. Or your best friend. Or your best friend who also happens to be a magician. It knows you inside and out, and shows you what you want — poof! — before you even knew you wanted it. So, how does Netflix come up with its recommendations for you, exactly? Well, it's a little complicated, but it's basically a two-part trick.
Up until December 2017, Netflix primarily used one algorithm, in particular, to dish out suggestions about what you should watch next, according to a collaborative article on Medium penned by the company's employees . For those not familiar, an algorithm is basically a set of instructions that tells your computer — or in this case, your Netflix app — what to do.
Algorithms are often used to facilitate machine learning (i.e. how your app picks up on your viewing habits), and "[instead] of repeatedly processing a stable set of instructions, systems [like Netflix] based on machine learning rewrite themselves as they work," as Slate explained in a 2016 article. So, every time you watch something on Netflix, they're collecting data that informs the algorithm. The more you watch, the more frequently the algorithm rewrites itself, and hopefully the better your recommendations get.
The data the streaming service collects to inform your recommendations is multi-faceted and a little complicated, but it involves way more than just the genre of a program. According to Todd Yellin, Netflix’s vice president of product innovation, they look at "what people watch, what they watch after, what they watch before, what they watched a year ago, what they’ve watched recently and what time of day," he told Wired in August 2017.
There are also tons of people at Netflix who apply all kinds of "tags" to each program — with everything from "how cerebral the piece is, to whether it has an ensemble cast, is set in space, or stars a corrupt cop," Wired noted — which even further informs your preferences.
Each of these data factors (and then some) carry different weight, but it all comes together to identify which "taste communities" you fit into. Single users can be part of multiple "taste communities," Yellin revealed, and how your screen gets populated — left, right, and top to bottom — is based on which communities you belong to.
After December 2017, though, the streaming giant upped their chances of serving you the right content, for the right moment, at the right time, by testing out a new recommendation feature based on artwork, the Netflix employees' Medium article explained. No, not the kind of artwork that you'd see at, say, the Louvre in Paris — the kind of artwork that you see attached to any given program on your screen.
This artwork experiment is the second part of Netflix's two-part recommendation trick. Even though it takes into account a lot of the same data factors as the first part of the recommendation factors, artwork is apparently even more important to users when it comes to deciding what to watch next, Nick Nelson, Netflix's Global Manager of Creative Services, revealed on the company's blog in 2016. Nelson explained,
"[Consumer] research studies ... indicated artwork was not only the biggest influencer to a member's decision to watch content, but it also constituted over 82% of their focus while browsing Netflix. We also saw that users spent an average of 1.8 seconds considering each title they were presented with while on Netflix."
Thrillist laid out an easy-to-understand example of how Netflix switches out program photos that cater to your previous watching history, using the movie Good Will Hunting as a starting point. Basically, if you see Good Will Hunting appear in your feed, Netflix is already assuming it's something you probably want to watch. However, the image your shown with the title is reflective of one of the reasons why they think you'll want to watch it.
"If you've watched a lot of love stories in the past, you might see an image of Matt Damon and Minnie Driver kissing," the article explained, "whereas if you're a comedy fan, you'll likely get a shot of Robin Williams."
Yes, OK, that makes sense. Historical romance fans would probably respond to a photo of kissing, and comedy fans would probably favor a still of a well-known comedian. Now, Good Will Hunting isn't really a comedy — although, there are definitely funny moments — and while there is a sub-plot romance, it's a drama in its entirety. It being a drama may have been why it was recommended to you in the first place, but the image you get should be reflective of those secondary genre-leanings — or the next best "taste community" you belong to.
So, there you have it. Netflix serves your recommendations to you based on a two-part trick, both of which have to do with algorithms and machine-learning, and one that anticipates which image is most likely to quickly catch your eye. It's an ever-changing, ever-advancing process though, so your recommendations are probably getting even better as we speak. Now go forth and find your new favorite show.