Hands on with Facebook Graph Search: Interesting, but disappointing
When I first heard about Facebook Graph Search, I expressed some doubts about how useful the feature would be for churning out practical results. Now that I have access to the feature, those doubts about whether most Facebook users will get much use out of it have only intensified.
To be fair, Graph Search remains in beta, and when announcing the service last week, Facebook indicated that a lot of fine tuning lies ahead. Still, Graph Search's heavy reliance on Likes may ultimately limit just how much information you can get out of your searches.
How it works
You can sign up for early access to the beta, but keep in mind that Facebook is rolling this out slowly. Should you make Graph Search active for your account, it will replace much of the header bar on Facebook. You click on the "Search for people, places and things" text at the top, and start typing in your search terms. Facebook provides plenty of real-time suggestions as you do, though not quite as intuitively as popular search engines like Google and Bing.
That's not too surprising: Graph Search is a very different take on search than what you're probably used to. It's not about keywords or any of the kinds of searches you'd typically perform. Rather, Graph Search focuses on token-based searches: Facebook recognizes key words in your search, and looks for data it has collected that matches what you're looking for.
That's easier to explain with a series of examples. When I type my friends into Graph Search, Facebook offers up a bunch of suggestions: Friends of my friends, Friends of my friends who live in Wyomissing, Pennsylvania (my hometown, though not where I live now), Friends of my friends who work at Macworld, Friends of me and my friends, and Friends of my friends from Wyomissing, Pennsylvania. (Yes, those suggestions are ever so slight variations on one another--almost to the point of annoyance.) I can click or use the keyboard to select any of those, use my original search, or customize my own search further.
Suppose I choose Friends of my friends who live in Wyomissing, Pennsylvania--perhaps because I want to find old high school buddies I haven't reconnected with on Facebook yet. I get a list of results, which I can instead choose to view in a grid layout. What I don't get, but could be hugely useful for this type of search, is an option to sort results, perhaps starting with people with whom I have the most friends in common first. Another bizarre omission: I can't find an obvious way to exclude my own friends from results. When I'm searching for friends of my friends, it's a little silly (and recursive) when Facebook includes people I'm already friends with in my results.
I can refine the search in all sorts of ways that range from creepy to clever--gender, relationship status, current location, hometown, year of graduation, and plenty more.
But let's move beyond searching just for new friends.
Data-driven suggestions and results
Facebook's suggestions when I type in Music my friends like include Music pages my friends like, Music places my friends like, Music liked by me my friends listen to, and Music my friends who like Unprofessional listen to. (Unprofessional is a podcast I co-host and have clicked to "Like" previously on Facebook--Likes figure heavily into Graph Search's results and options.)
Some of those suggestions get nutty, for different reasons. Any search you perform for music you already like--and there are oodles of possibilities--will, of course, return only bands you've already clicked to Like on Facebook. So you can see, I suppose, different cross-sections of people who share your tastes, but I'm not sure how useful that is.
Even more flummoxing is Graph Search's habit of offering up search suggestions with no results. I encountered those far too often.
Potentially more useful is the ability to see other interests people you explicitly or potentially trust have, that might prove interesting to you as well. For example: Music my friends like, or Music my friends who like "Weird Al" Yankovic like. You can go broader, too: Music liked by people who like Jonathan Coulton.
It's unclear how these results are sorted, too. I can't tell if the fact that Michael Jackson is listed first under Music liked by people who like Microsoft means something significant, compared to the fact that Rihanna tops the list of Music liked by people who like Apple.
Facebook lets you combine a lot of search factors. I now know that people who like both Rihanna and "Weird Al" Yankovic like Flo Rida, Adam Sandler, and Will Ferrell. People who like Will Ferrell and They Might Be Giants and Donovan McNabb seem to enjoy the Bible, Harry Potter, and Green Eggs and Ham. Atheists in Nebraska love Nine Inch Nails, Nirvana, and Johnny Cash. Women who like the drink V8 and live in New Jersey like television shows such as Cake Boss, The Ellen DeGeneres Show, and Law & Order: SVU.
Slice and dice but then think twice
No matter what search you perform, Facebook will offer options to hone your results with further refinement, or extend it: If you look for TV shows, you can also see books or photos the same demographic enjoys. It's a researcher's dream!
Or is it?
I found that the same results popped up again and again: A lot of people tell Facebook that they like the Bible, and many women have said that they like Ellen DeGeneres's program. And I suppose it's interesting to know that "Weird Al" Yankovic's fans like authors as diverse as Seuss and Poe, but it's not always especially actionable data.
Every time I thought I'd figured out a search that could maybe lead to unique suggestions that would genuinely interest me, Facebook let me down: Authors liked by my friends who like Dave Barry yielded a single result: Dave Barry. Turns out, that's because just one friend on my list ever clicked to like that particular humorist; though I suspect more like him, my graph results can only ever be as useful as the data my friends--and the rest of Facebook's citizenry--feed into it.
If I broaden that search to Authors liked by people who like Dave Barry, I get results such as David Sedaris, Dave Barry (again), and Douglas Adams--all good results, but not the kind of aha moment I was hoping for. Admittedly, that's precisely the kind of result that would seem most difficult for Facebook to provide, but still, I return to the question: What use is this?
Perhaps the most useful data comes in the form of recommendations for local establishments. (Annoyingly, Facebook isn't as smart as, say, the Siri personal voice assistant for the iPhone; you can't search for Restaurants near me that my friends like; you need to use Restaurants in Punxsutawney, PA that my friends like.) This gives you a private Yelp of sorts, with results as curated by your friends or by other locals, depending on how you search. The problem is, results are skewed by several factors: Restaurants that focus on social media is a big one; in my hometown, those establishments that maintain active Facebook pages score far more Likes than those that don't, though that doesn't necessarily reflect their quality. And similarly, only restaurants that are frequented by the kind of folks who think to seek out Like buttons in the first place get surfaced well in results.
So the data is interesting, but not necessarily meaningful if your goal is to find the best places you haven't heard of yet. A lot of people in my town like TGI Friday's. I didn't need Facebook to tell me that.
Despite my complaints, Graph Search is fun to play with. But that's just it: It feels like a toy. I quickly got good at predicting what TV shows folks who liked any mix of celebrities would like, because certain shows do quite well with active Facebookers, shows you can probably guess, with stars along the lines of Jon Stewart and Stephen Colbert.
Thus, even as a toy, it's hard for Graph Search to hold my interest for very long. I played with it for a few minutes at a time, and I suspect that that's what most people will do. The law of very large social networks, apparently, dictates that data aggregated from the masses isn't that interesting, and data from your own friends is unfortunately spartan. If Facebook can get me and you and everyone we know to provide more Likes on more interesting topics, Graph Search will get a whole lot more interesting.