Pinterest is massive. The visual scrapbooking site now hosts some 30 billion pins, nearly half of which, it says, were created in the last six months alone. In that vast storehouse of content, behind the DIY curtains and beneath the impossibly photogenic grilled cheese, there's almost certainly something you'd find interesting, but finding it has become a needle-in-a-haystack problem of a serious magnitude. You could scroll through Pinterest's rich front page feed for hours without coming across the things you're most interested in.
Pinterest's answer? A new type of search, one that cleverly taps the site's prodigious supply of user-provided metadata to help people find the things they're looking for--and in some cases, things they didn't know they were looking for to begin with.
Guided search, as it's called, arrives on Pinterest's mobile apps today. Its both hugely powerful and exceedingly simple. You start with a general query--say, furniture. Below the search field, a row of buttons pop up representing subcategories, algorithmically culled from captions, comments, and other metadata attached to pins satisfying your initial search. In the case of "furniture," the list might include things like "living room," "bedroom," "cheap," "Ikea," "modern," "do-it-yourself," and so on. From there, you can tap a subcategory to refine your search, at which point another set of sub-sub-categories emerge, giving you a chance to further narrow your results. At any point, you can back up to try a new path, or eliminate a particular filter from higher up in the chain.
The result is a search experience that's more like wandering toward an answer than being served one straight away. Ben Silbermann, Pinterest's CEO and co-founder, says the feature was designed to facilitate a more flexible type of discovery, a search where users could "refine, expand, or pivot" their query at any time. "We want it to be about showing you different possibilities about where you can go, instead of giving you a terminal answer," he says.
The approach is well-suited for a few different kinds of searches, many of which Google isn't especially good at. For one, guided search lets you start with very broad questions and amble toward a satisfying result. If you wanted to do something unique for your kid's birthday party, you could just type in "birthday," and peruse visual results around not only presents but cakes, crafts, decorations, and other activities.
It's also potentially useful for more specific, subjective questions, like figuring out what sort of haircut you should get. These types of searches, where every user's right answer will be different, are an awkward fit for Google's rank-ordered lists. Here, though, you might be able to narrow your results to just those that apply to your hair type.
In addition to these hazier types of queries, the new guided search could prove handy when you have something in mind but don't know the words for it. By winnowing down your results visually, modifier by modifier, Pinterest's guided search could take you to "mid-century modern chairs," for instance, even if you didn't have any clue that's what they were called.
What makes guided search possible is metadata--the user-supplied board titles, captions, and comments that cling to all those pins. No individual pin is filed as "living room furniture, Ikea, do-it-yourself" when it's introduced into the Pinterest system. Instead, those classifications emerge from the words people use to describe the pin itself.
Pinterest's search team, which includes veterans from Google and other big-name engines, built a system that looks for these sorts of descriptors, and, when they show up with enough frequency, turns them into the categories (and sub-categories) that accompany search results. The guided search categories aren't curated by Pinterest, in other words. They're curated by the millions of people who pin and re-pin on the site--in some sense unknowingly so. When you note in a caption that a particular hair style would be "perfect for thick, medium-length hair," you're helping add a bit of order to that messy ocean of pins. Just like Google's PageRank algorithm didn't look at websites so much as the links pointing to them, Pinterest isn't looking at images but rather the way those images have been described.
From this perspective, you could see guided search as a huge-scale experiment in crowd intelligence. By encouraging users to caption and comment, Pinterest has essentially enlisted its users in a Mechanical Turk-style exercise to put images--and their feelings about them--into words. The approach isn't without limitations; a search for "architecture," for instance, is more likely to yield "awesome" as a subcategory than, say, "Bauhaus." Still, this sort of semantic mapping offers some useful new ways of navigating Pinterest's singular data set.
Guided search signposts a slow, gradual shift in Pinterest's broader mission. Where it was once about collecting, Pinterest now focused is helping people discover, even when they can't articulate what it is exactly they're trying to find. "To us, that's the big unsolved problem on the internet right now," Silbermann says. "How do you discover things you didn't know you were looking for?"
It's a problem Silbermann's users are inevitably staring in the face. Pinterest's innovative staggered grid layout and infinitely-scrolling feed helped the site build a huge number of dedicated users. With astonishing speed, those users brought in images from around of the web, creating a rich pool of content within the site.
As that pool grows, however, Pinterest's main feed will only let people skim a vanishingly thin layer of its surface. With guided search, in addition to new tools like map-based pins and custom categories, we see a new focus on helping people dive deeper into the Pinterest ecosystem. You could say it's a shift away from letting users "happen across" things to a slightly more active model of engagement. Or just an attempt to create some algorithmic serendipity so people happen across things more often.
As Silbermann sees it, though, this matter of discovery is a pressing one not just for Pinterest but for the web as a whole. In spite of all the new platforms and services that have emerged in recent years, he points out, we're still half-stuck in a stone age of billboards and catalogs.
"It's crazy! People still ship a paper book and say, 'hey, you might find something cool here.' That's still the way you discover things. If you compare that to how far we've come in information retrieval, we just haven't made commensurate progress," he says.
Pinterest has the cool stuff, and they have the audience. The challenge now is figuring out how to best let people browse a catalog with 30-billion things in it.
