It’s the eve of Facebook’s IPO and everyone, and I mean everyone, is talking about it.
The nineteen year-old barista at my local Starbucks asked me a few days ago if I was going to buy. When I told him that I’d heard the IPO might be oversubscribed, that General Motors’ pulling its paid ads off of Facebook was a cause for concern, and that Google’s Knowledge Graph (from a revenue perspective) could give the social networking site a run for its money, he looked at me as if I had just broken his heart.
His slightly older manager rubbed his back for a minute then looked at me and said, “He wasn’t asking about all of that, he just wanted to know if you were cool.”
As I walked home, I got to thinking about what Facebook’s value really is (NOTE: I am not offering you, nor am I licensed to offer investment advice to anyone), whether its social graph and algorithms generate better recommendations and/or more hits than sites like Amazon, Netflix, LinkedIn, Match.com, Foursquare and Eventbrite, and if the company’s value can live up to its expectations over the long haul.
Though savvy entrepreneur and investor John Bortwick of Betaworks is bullish on Facebook, he talks about how quickly social, mobile and tablets have come into the forefront and how today’s vectors of destruction are so extreme. A lot has changed in a few short years.
Back in the Day
I was reminded of a history lesson Vipul Sharma, principal software engineer and engineering manager at Eventbrite gave me earlier this year. It’s worth sharing and it’s timely.
Sharma started way back in ancient times (the beginning of this millennium, remember Y2K?) when Amazon started making recommendations with the simplest logic. “If you were buying a camera, it would suggest that you might want to buy batteries for that camera too,” he explained. If it sounds primitive now, but it was somewhat revolutionary at the time. A computer smart enough to do suggestive selling.
Fast forward to 2003, Amazon had collected enough data to mine on top of its users. Bang! Item-based collaborative filtering was born. “Users who bought x also bought y,” explains Sharma. For the geeks among us, Amazon’s innovation works in two steps:
- Build an item-item matrix determining relationships between pairs of items
- Using the matrix, and the data on the current user, infer his taste.
Most enterprises, by the way, can’t yet do this very quickly, according to David Jonker, SAP Director of Product Marketing, Data Management and Analytics.
Next came Netflix’s algorithm which is based on user ratings and recommends similars. “People who liked the Godfather will probably also like Scarface,” explains Sarma. Geeks who want to know more about how this works should check-out the Netflix blog and the Netflix API.
Social Influence vs. Reading Patterns
After 2006, sites like Facebook opted to look at social connections vs. relying on patterns of buying. “Whatever your 'friends' like, you like too, is the idea,” explains Sharma. The social networking site mines the global mapping of everybody and how they're related, it doesn’t look at user:uses (Amazon’s earliest algorithm) or collaborative filtering.
“That was great at first,” says Sharma. But, in some cases, Facebook’s social graph might have become too dense to render the best placement of ads/recommendations.
That seems hard to believe until you consider your own social graph on Facebook. After all, it might include not only your school chums and fellow PTA members, but also your gardener, the rock star whose site you “liked” and your grandmother. Do you all belong in the same target market from an advertiser’s point of view?
The obvious answer is, probably not.
And while companies like Facebook will do all they can to train their customers (advertisers) to use their site better, new websites and services will look at new ways to gather and mine the world’s data with smarter formulas and algorithms.
What's the Next Model?
Take Eventbrite as an example. Its website allows event organizers to plan, set up ticket sales and promote events of any size and publicize them across Facebook, Twitter and other social-networking tools directly from the site's interface.
It looks at your explicit social graph (people who you are actually friends with or are familiar with, for instance your Facebook or LinkedIn connections) and your implicit (or interest) graph (people who are similar to you, as defined by your interests, but who may not be on your Facebook or on your Twitter) and combines them to provide relevant event recommendations.
Sarma says that his team mines a massive social graph of 18 million users and 6 billion first degree connections which requires processing more than 2,000,000,000,000 bytes (2 terabytes) of social graph information daily. All of that to recommend that I might want to take a Divine Dessert Class.
Want to see a picture of what I’d get to make?
I don’t know if I would need a social graph, an interest graph or a look into your Amazon or Netflix data, to predict whether I’d cause you to disengage with that last question. (Maybe all of the above, plus Pinterest?) The answer of course doesn’t matter. It’s your data that provides such great value and so many limitless opportunities.