The History Behind Facebook’s Graph Search

Editor’s note: Taylor Buley is a senior developer at Conde Nast’s PARADE. He’s a former staff writer at Forbes and graduated from University of Pennsylvania and Stanford. Follow him on Twitter @taylorbuley.

On Thursday Lars Rasmussen, Google Maps co-inventor turned Facebook Graph Search guru, took to Reddit for an “ask me anything” open thread. The Danish native avoided questions about the competitive landscape for Graph Search but spilled a near complete history of its development inside Facebook.

The Facebook engineer had a good time doing it, too, judging by the 18 smiley faces he riddled throughout.

Graph Search is Facebook’s foray into the search market. Instead of matching pages to search terms like “San Francisco + sushi restaurant,” Graph Search instead takes natural-language sentences like “my friends who like sushi” and finds results expressed through your social network. Facebook is betting that by using personalized data, they can provide more relevant search results than can mechanisms such as Yelp reviews or Google Page Rank.

Rasmussen writes that he was interviewed by Facebook in late 2010, around the same time Google announced the shutdown of Google Wave, a product launched by Lars and brother Jens. But it wasn’t until a half-year later that he was pulled onto the Graph Search project.

“Zuck asked me to work on search in the late spring of 2011,” he writes, recounting the first of three walks with Mark Zuckerberg. The Facebook founder “had a very strong vision for what he wanted and how compelling a structured search product over the content people have shared on Facebook could be.”

In another answer regarding the timing of releases, he explained how his team “showed the original prototype of what we much later named Graph Search in the early summer of 2011.” This prototype only took a few weeks to build, he said, but the project did source code from “previous prototypes of structured search products that were not based on natural language.”

Thus we know Facebook had played with the idea of a non-natural-language search product at least some point before the summer of 2011. This lends credence to rumors that circulated in 2010 of a search project built atop the freely indexable Open Graph tags standard it launched in summer of 2010.

What held up Graph Search development between the early prototypes and the January 2013 launch? Too many smart people at Facebook, perhaps. In a question regarding the “best and worst” of working at Facebook, Rasmussen discusses the pitfalls of having a company “chock full of passionate, brilliant, opinionated people.” The problem? “Sometimes it takes longer than I’d like to arrive at an answer.

“I think it is fair to say the project took longer to get to the beta stage than I predicted when we started,” the engineer candidly confessed. “Pretty much all projects I have ever worked on have had this property :) Time flies when you are behind schedule!”

But perhaps the delay was for the better. A vanilla search engine built atop Open Graph tags would have done little to innovate in the search space. Instead, in January Facebook rolled out a novel breed of search powered by natural-language queries and social data at a scale not available to, or indexable by, any competitor.

So what about Graph Search’s competitors? Members of the press have fingered Yelp and Google, among others, as possible competitors against Zuckerberg’s search vision. But when asked which companies Rasmussen and his team view as direct competitors, the engineer held back and merely proffered a smiley face.

Facebook Graph Search has yet to roll out to all users, and Rasmussen writes that part of the reason for the partial rollout so far was to allow live A/B testing on real users — a process that circumvents the possibility of endless internal discussions. “Without live usage we’d just be arguing all day,” he writes.

The company is apparently now discussing the future of Graph Search. According to Rasmussen, his third and latest walk with Zuckerberg came just last week and its purpose was to discuss the future of the search product.