One week ago, I received a Facebook message from my friend David. “SUPER random question, but do you know a girl from Michigan whose name is Lauren and lives in Boston?” he asked. “Apparently she is your friend and Jenna’s friend on Facebook.” I hadn’t spoken to David in years, so I was a little confused. “I was using Tinder (a new dating app) for the first time today,” he wrote, “and Lauren was the one person I decided to look up since we had 30 mutual friends and shared 30 interests. I guess we'll never know.” I was at work, and not in the mood to pore through the 20 Laurens in my friends list to help David find the right girl. Then I remembered I had been given access to Facebook’s new search product, Graph Search, which is built to answer such questions. I opened Facebook and typed into the search bar, “Friends of mine named Lauren who are also friends with Jenna and live in Boston.” Facebook returned one result. Bingo.
I have no idea how or if David tried to set up a date with Lauren, but even in "early beta" Graph Search had certainly proved its mettle at finding people. It felt like using Google for the first time, seeing tons of data scraped through in mere seconds. All of the information you have ever provided about yourself has been indexed and cross-referenced, accessible with a few keystrokes.
Facebook launched Graph Search at a big press event at its Menlo Park, CA headquarters almost exactly one month ago. CEO Mark Zuckerberg delivered a large part of the event keynote himself, highlighting the feature as one of three “pillars of Facebook” alongside the News Feed and Timeline. Graph Search is supposed to help you gather friends for a Twin Peaks marathon, find photos taken in London on your last trip, and see which sushi places are most popular among your friends. After a month of testing Graph Search, I’ve found that it’s fantastic at finding people and photos, but not so good at finding anything else. And it’s for one simple reason.
A currency of Likes
After my first few moments playing with Graph Search I immediately recalled Advanced Search, a feature Facebook dumped back in 2008. As a Freshman at the University of Michigan in late 2007, Advanced Search let me easily see which Freshman girls were interested in Scottish twee pop and Stanley Kubrick movies. No dates or hookups ensued, but I was nevertheless captivated by the ability to perform such granular searches for people by interest — drilling down by class year, gender, age, and even by where a person was from. Facebook generated dozens of results in response to any query I inputted. In part this was because at the time people, especially college kids, used to religiously fill out all their profiles with all their favorite bands, books, and movies. The social network had only been around for a couple years, and everyone was still giddy about having a "clean" place to post things about themselves online, since MySpace had become a messy junkyard of auto-playing music embeds and party promoter spam.
Regardless of how many people gain access to Graph Search over the coming months, these results will not change
In 2007, clicking the Starship Troopers link on my profile populated a list of other nerds who loved the movie. No Facebook landing pages existed for movies and interests. Today, however, clicking Starship Troopers takes you to a page operated by social media gurus at Sony Pictures Entertainment. Over time, Facebook turned over all band pages to the bands, and all movie pages to the studios. Before you knew it, posts from these pages turned up in the News Feed, and soon these posts evolved into "ads." This means Sony now fills me in on upcoming Starship Troopers sequels right in my News Feed, and when you’ve liked dozens of pages over the years, it all adds up. Likes have become so commoditized that I, a self-professed Facebook fanatic, have become stingier with them, and so have many of my friends. When I like Nike or a restaurant nearby, I am acutely conscious that I'm signing up for News Feed updates for life.
Fast forward to today, where Graph Search decides what’s important in large part based on Likes. If you search for "sushi my friends like in New York City," Facebook returns sushi restaurants friends have liked or checked in to. The restaurant recommendations from friends are few and far between, and in the case of sushi restaurants, are completely empty for me. Each restaurant recommendation is accompanied by the friend(s) who liked it, but since none of my friends have liked a single sushi restaurant in New York City, no results appear. Regardless of how many people gain access to Graph Search over the coming months, these results will not change. If you search for the more general "sushi restaurants in New York City," the results are better, but still not anywhere close to Google, Zagat, or Foursquare’s caliber of recommendations. The top suggestion is Kumo Sushi, a restaurant known not for its great sushi but for its $32 all-you-can-eat sushi and sake bombs deal. The deal has likely drawn a higher than usual number of young patrons and thus Facebook check-ins, which ranks it higher in Graph Search. It ranks higher than Nobu.
The restaurant recommendations will not improve until Facebook finds a way to incentivize users to check in, like, and rate restaurants. Users provide Facebook with plenty of data about where they live and who they’re friends with, but hardly any for stuff they like to do. Ironically, in its efforts to encourage restaurants to advertise on the site, Facebook unwittingly discouraged users from inputting the critical data Graph Search relies on. This conundrum is critical to the future of Graph Search. Without data from friends about what they like, Graph Search’s most overtly monetizable aspect is useless. But the game isn’t over yet.
A reason to Like
Graph Search Product Director Tom Stocky
Graph Search Product Director Tom Stocky says that there are a few tangible benefits to filling out your interests on your Facebook profile in the world of Graph Search. Most importantly, it’s fun to be "the guy" your friends go to for horror movie recommendations, or the girl your friends go to for dinner date ideas. People feel good when friends enjoy their recommendations, Stocky says, which will drive many users to list their favorite things on Facebook. "As soon as we had Graph Search internally, I found myself liking more stuff and checking in to more places," he says. "I was giving a gift to my friends, in some ways. If there's a reason it helps your friends, that might be a reason you share more." In Facebook’s Graph Search announcement, Graph Search Engineering Director Lars Rasmussen spoke about how he liked a dentist’s Facebook page in order to aid his friends.
While Likes may today form the backbone of Graph Search recommendation data, Stocky says data from Open Graph apps (like RunKeeper and Spotify) may someday be more important. Graph Search will have answers for which trails are most enjoyed by your friends who run using RunKeeper, and which songs are most popular among friends who use Spotify. Individual trails and songs don’t have pages, and thus can’t post things to your News Feed, which may entice more users to try automatically posting activities using those apps. "It could be really useful when you're throwing a party," Stocky says. "You could ask, ‘What's the favorite music of the people throwing this party?’ There are so many in-the-moment things like that." With Graph Search, Facebook is very clearly attempting to match what Google achieved with search ads: monetize intent.
"As soon as we had Graph Search internally, I found myself liking more stuff and checking in to more places."
Dirty data
The solution might lie partly in improving data analysis
But intent may not be as easy to measure as Facebook thinks. GroupeMe’s head of Business Development Steve Cheney hypothesized that a large portion of Facebook’s Likes are "dirty": they don’t mean anything, and are based on ad campaigns years ago that had the sole goal of acquiring Likes. The truth, he wrote on his blog, "is that the link between query intent and your social interactions for interests and places is much weaker than FB wants you to believe. As much as half of the links between objects and interests contained in FB are dirty — i.e. there is no true affinity between the like and the object or it’s stale," he wrote. "For the past several years big advertisers on FB have actually been directing massive amounts of paid media to acquire fans," he continued.
Cheney references paid ad campaigns whose goal was to earn a Like using a "Like to enter" or "Like to qualify" conditions. "One direct effect of all this passive liking is an ugly messy data set with a bunch of implicit signals… that are wrong," Cheney continues. The result might be that PF Chang’s is the most liked Chinese restaurant near you simply because it had a "Like our page to win some lo mein" promotion two years ago. "You simply can’t roll up recommendations for people, places, and interests into a service that’s one size fits all," Cheney concludes.
Cheney’s data may be completely off, but the gist of his argument seems to hold water. When I search for "things my friends in Ann Arbor, MI like," I first see the pages for Michigan Football, The Michigan Daily, and UMich Memes. Just below them are Good Time Charley’s (a great place to grab a beer in Ann Arbor) and Chase Community Giving, a charity page liked by a whopping 183 of my friends. The Chase page is ranked much higher than Sava’s, one of the most popular restaurants in Ann Arbor. It’s no stretch to guess that Chase at one point set up a contest that perhaps exchanged Likes for charity money — a noble goal — but now Facebook’s rankings for Ann Arbor are interminably skewed. Mark Zuckerberg admitted at the Graph Search launch event that the company needs to do more to encourage people to rate restaurants and to Like things, but convincing users skeptical of Facebook as a public company isn't going to be easy.
The solution might lie partly in improving data analysis, Stocky says. "If you want to say you went to Hogwarts, that's totally fine," Stocky says. "We want people to be able to express themselves in this way." He envisions a world where fellow Hogwarts grads might find each other through Graph Search. "A lot of people say they work at Facebook, even though they don’t," he says, "but we rank people higher who we have higher confidence actually work at Facebook. I don’t think of it as misinformation. I think of it as self expression." Graph Search is an "early beta product," and with time, its results could more closely correlate to what people actually like, and scrape out the bogus results.
Plenty of friends doesn't lead to plenty of data
When Graph Search rolls out to more of my Facebook friends, they might feel the need to publish more information about themselves to help others, but I suspect they won’t — because who has the time? Which brings us to Mark Zuckerberg’s biggest assumption: that Facebook users interact with the social network in a similar way to Facebook employees, who spend their lives on the site, use Facebook Groups to organize team meetings, and use Graph Search to organize ski trips and movie nights. Users don't necessarily realize that liking things and adding interests no longer correlates to News Feed posts the way it once did.
Facebook has improved its algorithms for displaying content in the News Feed, and even added a "hide" drop-down button so you can like a Page but not get updates from it. "If I go and like a few hundred local businesses but I'm not interested in their updates, the system is designed to automatically recognize that, and show me fewer updates from them," News Feed product manage Will Cathcart says. The challenge is educating users.
A line of code can’t fix the fact that none of my 150 friends in New York City have liked a single sushi restaurant on Facebook
With Graph Search, Facebook has laid the groundwork for a future of finding friends, photos, and stuff to buy and do. In fact, the finding friends and photos part works wonderfully, letting you dig into years of information by simply typing in a sentence. Facebook understands any way you want to search for your friends, whether they’re "homies," "besties," or even "photos of randos that went to my school in 2008." The site indexes 300 million photo additions per day, cross references them all with each user’s privacy settings, then populates it all in real time. But until Facebook users start liking more things organically, only half of Graph Search’s potential works as advertised.
There are dozens of cases where Graph Search comes in handy, and someday we may have trained ourselves to use those features. "Which of my friends speak French?" you might ask as you attempt to decode a menu at an authentic French restaurant. "Which of my friends have been to St. Petersberg?" you might ask as you look for ideas for an upcoming trip. "People named Jenny who went to Arizona and are friends with Dana," you might input after meeting someone at a party. Searching for "Photos of me in 2009" is much easier than digging back through your Timeline manually. You could call Graph Search's people and photo-finding capacities stalking, or you could call it researching. Its Facebook getting back to its Advanced Search roots. Whether queries like "best restaurants near me," on the other hand, will ever be as useful as a Foursquare or Yelp search is right now still an open question. Graph Search is by all means an enormous technical accomplishment, but its "killer feature" has yet to show its head.
Facebook calls Graph Search an "early beta," but the weird part is that it doesn’t feel like a beta product. As a recommendations search engine, Graph Search's engineering is sound, but there isn't enough data to fill it up. A line of code can’t fix the fact that none of my 150 friends in New York City have liked a single sushi restaurant on Facebook. "Every day, we get 2.7 billion likes, 2.5 billion status updates / photos, and 300 million photos," Stocky says. With all that great personal data, it seems like Facebook could be earning even more. The success of Graph Search will be defined not by how much effort Facebook puts into it, but by how much users put themselves into Facebook.