The Problem With Netflix’s Goofy Sub-Genre Algorithms


Because we love to hear about the secretly complicated ways in which the things we take for granted work, Alexis C. Madrigal’s recent Atlantic deep dive into Netflix’s ultra-specific genre-generation algorithm has been inspiring quite a bit of discussion. And for good reason — it’s a fun piece, a painstakingly researched (with said research painstakingly described) examination of one of the streaming service’s goofiest elements, complete with charts and graphs and a “Netflix-Genre Generator” and even a lengthy, on-the-record interview with the folks at Netflix (who aren’t always so open about the nuts and bolts of their organization). But its underlying assumptions and conclusions are a little dodgy, the result of a bit too much consumption of the Netflix Kool-Aid.

What’s strangest about Madrigal’s piece is that it seems to want to have it both ways — to have a good hearty laugh at those weirdly targeted, intricate sub-sub-genres via their jokey Generator and ill-advised wacky graphics from pretend movies (like Reservoir Cats, from the “Violent Thrillers About Cats from Age 8 to 10” grouping, ho ho), but at the same time, to look in wonder and awe at the sheer volume of those sub-genres, to crunch the data related to them, to piece together the formula of their creation, and then to visit Netflix HQ and geek out with the guy who creates them.

So, which is it? Are Netflix’s sub-genres a bold and innovative attempt to, as the headline promises, “reverse engineer Hollywood?” Or are they a laughable attempt to serve up hyper-targeted C-movies based on dubious connections to other things you’ve viewed?

This viewer leans towards the latter. Reading through this detailed examination of all the elements people “want” to see in movies, I couldn’t help but recall that article in the New York Times last May about Vinny Bruzzese, the statistics genius who was supposedly taking Hollywood by storm with his “script evaluation” services, analyzing screenplays and targeting the elements that would assist or cripple its chances at the box office. That smelled like bullshit, data miners and numbers crunchers taking entertainment apart, entirely free of context, to create the kind of soulless, slavish imitations that fill coffers but never change the game. This is the same thing, on the back end of the production equation — Hey, you like period romances set in Europe, right? Here are some other ones!

But the shallowest surface elements of a film aren’t always what draw us to them; it’s a (sorry) intangible mixture of those elements and reputation, reviews, personnel, mood, and convenience. (More on the latter point presently.) What Netflix does here is closer to Patton Oswalt’s bit about his first Tivo, which followed up his recording of the Western classic The Man from Laramie with a bunch of children’s programming set on farms. “But you like the horsey shows!” he imagined his Tivo reasoning, and more often than not, Netflix’s conclusions operate from the same logic.

These recommendation engines must be working, or else Netflix wouldn’t keep using them, so my anecdotal experience — that I’ve never used them for anything more than a giggle, and don’t know any serious cinephiles who’d take their recommendations seriously — is probably irrelevant. But I know this much: these intricate categorizations are basically filler for the Netflix front page, a clunky interface that pushes lesser titles so we don’t notice how infrequently they’re streaming what we’re actually looking for. Reuters’ Felix Salmon made this point shortly after the Atlantic piece went up, explaining that since Netflix “can’t afford the content that its subscribers most want to watch” (i.e., the biggest new releases and buzziest indies), it’s instead resorted to “scouring its own limited library for the films it thinks you’ll like.” The ultimate problem, Salmon notes, is that Netflix “doesn’t want to be movies: it wants to be TV.” We noted as much after last May’s “Streamageddon” debacle, as the company slyly pivoted from pushing their loads of content to a new mission: “Our goal is to be an expert programmer, offering a mix that delights our members, rather than trying to be a broad distributor.” And pushing the streaming library’s hyper-targeted, second-tier titles is “expert programmer” writ large.

To be clear: we like Netflix. Just last week, we highlighted notable streaming titles not once, but twice. Alas, that one-two punch was prompted by yet another of their increasingly frequent, giant changeovers in streaming content — and that’s part of the problem. The company put Blockbuster out of business (and that was no great loss), but with the power of owning movie rental comes responsibility, and they’ve fumbled the ball frequently, from indifference to older titles to aspect ratio issues to the disappearance of physical media. It’s not surprising they wanted to talk to the media about how their nutty sub-genres work; it’s a nice diversion from their many other shortcomings.