Face­book tries to build a search en­gine of its own

▶ It’s build­ing a tool that’s like a Twit­ter-google mashup ▶ “If they get it right … it’s ex­tremely lu­cra­tive”

Bloomberg Businessweek (North America) - - Contents - �Sarah Frier

Ev­ery few months, Twit­ter in­fu­ri­ates die-hard users by chang­ing it­self to work more like Face­book. Tom Stocky’s job at Face­book is to fig­ure out how to go the other way, build­ing a search func­tion that could make the so­cial net­work bet­ter at col­lect­ing the lat­est thoughts on Bernie San­ders’s

tax plan, or which Os­car win­ners are draw­ing chat­ter, or where in Thai­land to snorkel.

Face­book com­pletes 1.5 bil­lion search re­quests ev­ery day, but most are pretty sim­ple, like the name of the guy you just met in that bar. In Oc­to­ber, Stocky’s team qui­etly made it pos­si­ble to sift all pub­lic posts for re­sults on any kind of search, mean­ing you can re­search ho­tels or recipes with rec­om­men­da­tions from all 1.6 bil­lion users, not just your Face­book friends.

If that sounds like Google, the sales pitch is al­most word-for-word Twit­ter. “What we re­ally tried to do was make Face­book a place where you could tap into the global con­ver­sa­tion of what was hap­pen­ing in the world,” says Stocky, vice pres­i­dent for search. “We re­ally want to ba­si­cally make Face­book the best place to find what peo­ple are say­ing about some­thing right now.”

That’s not easy. Face­book’s search al­go­rithm has to as­sess and rank tril­lions of user posts—more Web pages than Google’s search en­gine in­dexes. And Face­book wasn’t par­tic­u­larly good at search even when it was much smaller. While a search bar has been sit­ting promi­nently atop the home page since shortly af­ter the site’s cre­ation in 2004, its re­sults didn’t ex­tend be­yond names of friends to con­nect with un­til 2010.

At that point, Face­book struck a deal with Mi­crosoft to show Bing re­sults in its search bar. In 2013, Face­book un­veiled its own at­tempt at a more gen­eral search tool, called Graph Search. It re­quired users to en­ter queries in a highly styl­ized way, as if they were talk­ing to a ro­bot: “My friends who went to high school in Long Is­land and live in Philadel­phia,” or “Google em­ploy­ees who like na­tional parks.” Not sur­pris­ingly, Graph Search never caught on, and Chief Ex­ec­u­tive Of­fi­cer Mark Zucker­berg later ad­mit­ted it worked as in­tended less than half the time.

In the past cou­ple years, Face­book has done more re­cruit­ing in the field, lur­ing search en­gi­neers from com­peti­tors and re­think­ing how the process should work. Search on Face­book, like any­where on­line, starts with in­dex­ing. The mo­mentt some­one pub­lishes a post, Face­book’sbook’s al­go­rithm scans the sen­tences for key­words that can help cat­e­go­rize it. It then makes more qual­i­ta­tive judg­ments, in­clud­ing some that are fairly spe­cific to Face­book. Is this well-writ­ten? Are there gram­mat­i­cal er­rors or mis­spellings? Is the writer an au­thor­ity on this sub­ject, based on past posts? Does she usu­ally re­ceive many likes and com­ments?

“We need to make sure that all the up­dates—ev­ery sin­gle photo, ev­ery sin­gle video, link, share, like, com­ment—are re­flected in our in­dex in a mat­ter of sec­onds,” says Girish Ku­mar, Face­book’s di­rec­tor of en­gi­neer­ing for search. Ku­mar says Face­book’s al­go­rithm pri­or­i­tizes re­sults from orig­i­nal sources and au­thor­i­ties. Friends will ap­pear high in the re­sults, but so will brands, pages, celebri­ties, and knowl­edge­able strangers.

Face­book has hired hun­dreds of em­ploy­ees and con­trac­tors to test the search soft­ware. As re­cently as a cou­ple of months ago, the re­sults weren’t that good, Stocky says. Get­ting recipes when search­ing for Turkey, the coun­try? Face­book didn’t ac­count for some con­tex­tual fac­tors—the phone mak­ing the search was in Istanbul.

Along with bug testers on the pay­roll, Face­book’s users are in­di­rectly pro­vid­ing data on the search tool’s util­ity. When you click on a post or share an ar­ti­cle, up pops a link that shows how many peo­ple are talk­ing about the item and sug­gests re­lated sto­ries. When you use the so­cial net­work to log an ap­pear­ance at a restau­rant, Face­book Search de­liv­ers re­views from other vis­i­tors. “If search on Face­book re­ally takes off, I think Google is un­der pres­sure,” says Vic­tor An­thony, an an­a­lyst at Ax­iom Cap­i­tal Man­age­ment. “If they get it right and they’re able to mon­e­tize against searches, it’s ex­tremely lu­cra­tive.”

It’s un­clear whether most peo­ple will want to use Face­book as a kind of uni­ver­sal search en­gine, says Mark Ma­haney, an an­a­lyst at RBC Cap­i­tal Mar­kets. “They think of Face­book as the place they com­mu­ni­cate and share with their friends.” The com­pany “would have to come up with a bet­ter search ex­pe­ri­ence to get peo­ple to con­sid­er­con­side us­ing their search func­tion­al­ity in­stead of Google’s. The odds are very slim.”

Stocky says he’s used the search tool to get baby-mon­i­tor rec­om­men­da­tions from friends who’ve posted about cer­tain brands. For now, though, he’s work­ing on broad­en­ing the con­ver­sa­tion. “You should be able to tap into th­ese per­spec­tives and ex­pe­ri­ences from peo­ple you don’t know,” he says. “When peo­ple ac­tu­ally as­so­ciate Face­book with an­swer­ing the ques­tions they have, that’s when we’ll be suc­cess­ful.”

“We need to make sure that all the up­dates—ev­ery sin­gle photo, ev­ery sin­gle video, link, share, like, com­ment—are re­flected in our in­dex in a mat­ter of sec­onds.” —— Face­book’s Girish Ku­mar

The bot­tom line Face­book is work­ing on ways to im­prove the search­a­bil­ity of its tril­lions of posts and make its net­work more of a real-time re­source.

Newspapers in English

Newspapers from Canada

© PressReader. All rights reserved.