Stay on Top of Enterprise Technology Trends
Get updates impacting your industry from our GigaOm Research Community
Facebook (s fb) engineers posted more details Wednesday on the back end of its Graph Search function, showing how the social network assigns numbers to users, places and other reference points and then lets users form queries and find answers using those numbers.
Unicorn, the software and search engine that makes Graph Search possible for at least hundreds of thousands of users, starts by giving every user a number. In the example cited by Facebook Engineering in a blog post, a fictitious person named David has a number, or fbid, of 10003. His home, New York, is 111. And “Downton Abbey,” a television show David has liked on Facebook, is 222. Friends of David get called up with the search term “friend:10003.” People who live in New York are at “lives-in:111,” and people who like “Downton Abbey” live at “like:222.” Put those three strings together, and you’ll get other friends of David who live in New York and like “Downton Abbey.”
In each search string, sequence is important. The post states that Unicorn serves up quick results by quantifying the importance of each element of a search string and then sequencing those elements in order of importance. But at a whiteboard session last month at Facebook headquarters in Menlo Park, Calif., Facebook engineers said they want to automate the process of flipping around users’ search strings to trigger better search results. It’s one of a handful of things the engineers are looking to do to further improve Graph Search in order to live up to the company’s lofty goals for it, as I reported after attending the whiteboard session.
New likes per day alone number more than 2.7 billion, according to Wednesday’s blog post. At that rate, the number of possible fbids clearly will continue to grow, and the search strings will get longer, too. Turning out good search results in a couple of seconds could become more of a challenge.
It’s a good thing Facebook is innovating on the hardware side through the Open Compute Project. That work could become a higher priority if Facebook grows at a faster clip, although at least a few users are quitting for a slew of reasons.
Wednesday’s post does not mention advancements on Graph Search since the whiteboard session, even though the engineers said they would improve Graph Search in the months ahead.
Facebook engineers work with big data sets in several other ways, often with Hadoop. Facebook’s engineering manager of analytics infrastructure, Ravi Murthy, will moderate a panel on the future of Hadoop and business intelligence at GigaOM’s Structure:Data conference in New York in two weeks.