The Limitations of Facebook’s New Search Tool

As I reported in Monday’s New York Times, Facebook is rolling out its graph search feature on Monday to all of its English-language users in the United States.

In theory, the tool, which has been undergoing refinement since public testing began in January, will allow users to search through their social network to find friends or other connections with a common link â€" say, “women who like Mitt Romney and Nancy Pelosi and live in San Francisco” (Facebook reports there are 12) or “my friends who like hiking.”

In practice, graph search has many limitations. Because Facebook uses so-called structured search technology to scan the boxes and buttons that users fill out on the site, it’s only as good as the boxes checked.

For example, when I searched my social network for “friends who like ice cream,” Facebook only returned 12 results. Left out was the silent majority who regularly lick cones and never bothered to tell Facebook.

Graph search also tends to be too literal, often failing to discern the broader meaning of specific data points. For example, a search for “my friends who work at newspapers” returned no results, even though several hundred of my Facebook friends have listed specific newspapers as employers.

In short, Facebook’s graph search lacks the intuitiveness that people increasingly expect â€" and often get â€" from Web search engines like Google.

“You can search for restaurants my friends like, but ‘like’ is still binary on Facebook â€" they hit the like button or not,” said Nate Elliott, a principal analyst at Forrester Research who studies social media. “What if I want to find restaurants my friends love on Facebook. Or ones that get at least 3 out 5 stars?”

“There’s a lot of nuance required to make this really useful,” he said.

Over time, Facebook plans to encourage people to share more data to make graph search more valuable.

When I was on my profile page over the weekend, for example, the site prompted me to like more television shows, pulling up a long list of suggested titles, from current hits like “Game of Thrones” to moldy oldies like “Benson,” a political comedy popular in the 1980s.

The company is also considering how and when to prompt people to check in more frequently and write reviews without being too pushy about it.

“After you check in, half an hour later, do you want to rate that restaurant?” said Loren Cheng, who leads the team of linguists working on graph search. Or would you prefer to be prompted during dessert? Facebook is trying to figure that out.

Mr. Cheng said users’ own search experiences will be the biggest force prompting them to share more information.

“In user experience studies of likes, they say ‘why does it matter that I like something, why does it matter if I check into some place?’ ” he said. “You almost have to get them to a search result and they say, ‘wow, that’s what can I get. Tell me what I can do to get better results.’ ”

The search feature will also get more powerful later this year, when Facebook intends to add the ability to search through the text of status updates as well as information posted to Facebook by some third-party applications, like hotel and restaurant review sites and the streaming music service Spotify.

Consumer brands are already scrambling to tune their pages for graph search, said Susan Etlinger, an industry analyst with the Altimeter Group.

For example, big hotel chains like Marriott want to make sure that every property has its own Facebook page â€" and therefore could be easily liked by customers and found in a future graph search by their friends.

“Down the line, when people are searching for a hotel, you want to show up as a hotel with a positive reputation,” Ms. Etlinger said. “The experience you have at a Marriott in Minneapolis may not correspond to your experience in Houston.”

Facebook’s partnership with Bing, Microsoft’s search engine for the broader Web, will also benefit from better graph search, since Bing users who choose to link that engine with their Facebook account can pull up results from their social network along with overall Web results.

“When I look at the potential to create a more unified experience, they are making a very smart play,” Ms. Etlinger said.