The foundation of the company's pitch is quite simple: CEO Amit Kapur says Gravity knows what you like and what you want better than any search engine. That information, of course, is gold to advertisers.
How Gravity gets that data and the theories behind it are a bit more complex than the simple pitch. Gravity has created an ontology of interests. Underneath all the company's data-mining technology is a database of about 4,000 concepts that are connected to 7.5 million "interests" and 100 million colloquial phrases. Human "ontology monkeys" (my term, not Kapur's), algorithms, and eventually crowdsourcing will keep the ontology up to date.
When you use one of these 100 million phrases in a social update--a Facebook post, a tweet, or a blog entry--Gravity notes which concepts you're talking about and files them under your name. Likewise, when you include a link to a Web page in an update, Gravity divines the top concepts from that page and those go in your profile too.
(The company's previous product, Conv.io, was a "social take on forums," but when Twitter and Facebook opened up to developers, Gravity changed its direction to start using those feeds instead.)
If Gravity works as advertised at profiling users' interests, it should be able to make some nice money from the advertising business. But there's more to the company's vision. Gravity is going to connect people not just to their interests, but to other people with compatible interests. As Kumar says, "We're able to tell you not which people you may know, but which people you should know."
This data can be used to make better friend recommendations on Facebook. And it could be a killer service for a dating site like OkCupid. But, again, the real money here is advertising. In its database, Gravity can connect users to the people most likely to influence them. With another turn of the crank, it can find the most influential people for a given category of goods or services.
The advertising market Gravity is chasing here is not the standard pay-for-performance or click-through model that is already successful on the Web. Kapur says that while 33 percent of that direct advertising market is online, only 6 percent of the more-interesting and harder-to-measure brand advertising market is. With Gravity technology, he says, advertisers can target the "right audience" of influencers and measure the resultant uplift in awareness and favorability.
Developers will be able to tap into the Gravity data through APIs.
As Gravity works to give its advertising business some lift, it will roll out what it hopes are fun and useful consumer products. These services may help the medicine go down, should some Web users start to worry about the dossiers that Gravity is building on them. Twinterest, for example, lets you compare your interests on Twitter with other users. An upcoming personal newspaper service, The Orbit, will find stories for you based on your activity on the service as well as what you write about on social sites.
As far as security, Kapur says his service will never publicly match interests to individuals, nor sell names to advertisers. The APIs will deliver ads where clients want them, but the users won't get delivered back. However, Kapur does acknowledge that keeping user identities walled-off will require constant monitoring and adjustments.
And there is one data set that does not feed into the Gravity service: search behavior and browsing. "That would seem invasive," Kapur says. Instead the service only collates what you actively share in the public sphere. Right now that means Facebook and Twitter; but later Gravity may add Foursquare, Yelp, and other social sites.
It wasn't clear to me how Gravity will work with the lurkers of the Web: People who consume information but don't share it publicly--those with no Twitter account or locked down, family-only Facebook settings.
Still, Gravity is ambitious. Its technology is a potential threat (or an opportunity, depending on how you see it) to the advertising platforms at both Google and Facebook. It's also one of the most interesting machine learning and data mining experiments I've seen.