Netflix’s genius movie recommendations come from an army of “taggers” whose job is to sit around watching movies all day, then describe them in extremely detailed terms. Here’s how they do it.
In countless forums scattered across the internet, Netflix’s 25 million members prove week after week that the on-demand movie and TV provider is basically the online equivalent of a fortune cookie. They peer over each other’s digital shoulders as they compare, brag, laugh and ask each other, “What did you get?”
Except they’re not talking about fortunes. They’re talking about another form of prediction entirely: Netflix recommendations. From “Gory nightmare-vacation horror movies” to “Inspiring fight-the-system movies based on real life,” Netflix’s endless collection of Final Jeopardy-worthy sub-genres has sparked an online culture of one-upmanship and, yes, even a satirical Netflix Sub-Genre Creator.
The categories sound laughably long-winded, but they’re also frighteningly accurate. This comes down to Netflix’s algorithms— the intricate sequences of numbers, codes and calculations that serve us with content and ensure we remain entertained. And behind those algorithms is Todd Yellin, Netflix’s vice president of product innovation, and a former filmmaker who’s proven his own chops: he’s done British TV docs, smuggled footage about political oppression out of Burma, and was the first person to film Tibetan refugees escaping over the Himalayas.
He’s also the man who can read your mind. Here’s how he does it.
1. Netflix is secretly a matchmaking service
Like it would on any online dating website, a new account on Netflix begins with a questionnaire. The site wants to know whether you love steamy rom-coms or low-budget slashers, and it matches you up appropriately. So it’s only fitting that Yellin describes the process of determining viewers’ recommendations as a “marriage” of art and science. “We really have to understand our users,” he says, “so we consume vast amounts of data about what they play, how much they play and on what devices they play.” While the first few dates with Netflix’s recommendations might be rough, the system gets to know you by monitoring the movies you watch and how you rate them. Yellin says one user might became so fond of the algorithm, they actually tweeted saying, “Can I date my Netflix account? It knows me.”
2. What you think you want to watch isn’t what you really want to watch
“There are two versions of any person,” explains Yellin. “There’s the actual person and then there’s the aspirational version.” It’s a fine balance between giving a viewer what they say they want in the questionnaire and recommending what their behaviour has proven they actually enjoy. “You might want to watch environmental documentaries, but tonight you’re going kick back and watch Date Night with Steve Carell for the third time.” As you view content, Netflix gradually does more as you do and less as you say, until your viewing behaviour ends up determining about 60 per cent of the real estate in your recommendation feed, while suggestions to humour your aspirational self occupy the remainder (like a good partner, it hasn’t quite given up on your idealistic goals).
Netflix also has an aspirational version of itself. “Our utopian state is that when you fire up Netflix, we won’t show the thousands of movies we have,” says Yellin. “We show one movie or TV show and that’s it. And it’s exactly what you want to watch at that moment and you hit play and you’re done.” And if that movie just happens to be An Inconvenient Truth, so be it.
3. Netflix has an army of taggers working around-the-clock
With all the talk of algorithms handling the heavy lifting, it’s almost comforting to know that the real legwork of the personalization process can’t be done by computers (at least, not yet). Instead, Netflix employs about 40 freelance “taggers” across the globe, whose job it is to watch movies and TV shows and classify them according to more than 100 different data points, from as obvious as genre, director and starring actors to as obscure as the ease of viewing and the strength of the female characters.
“Tagging is really about being able to dissect a film into specific details but also understand the big picture of it,” says Toronto-based tagger Jordan Canning, who’s one of only a handful in Canada. “It’s balancing those out and being as concise and as accurate as you can be.” To achieve that balance, Canning sits down with a tagging “spreadsheet” and watches the films she’s been assigned—typically three to five a week—marking it up as she goes. It can take her up to an hour after watching the movie to complete the tagging process, and then the machines take over.
When she’s not tagging for Netflix, Canning is making her own films or supervising scripts, which is common among the company’s tagging community. “We have people who are screenwriters in L.A., film critics, a theatre critic,” says Yellin. “They could be tagging these movies at midnight or at nine in the morning. We don’t care.”
4. Movies with talking animals are a pain
All movies and TV shows are equal when it comes to tagging, but some are more equal than others. While kids’ shows are generally pretty easy (you can assume there won’t be any profanity, nudity or drug use), others are harder to classify within a set list of storyline and tone tags. “One of the hardest films I ever had to tag was a really long historical drama in German,” says Canning. “It had dozens of characters and lots of sex in it, surprisingly… or maybe not surprisingly.”
And while foreign films are a consistent challenge, so are talking animals: “One of the categories in the character section is occupation. Sometimes you just have to decide whether a fox can have an occupation.”