Know Thy User! GraphDive Distills Inferred Demographic and Interest Graph Data from Social Feeds

By Michael Carney , written on October 4, 2012

From The News Desk

Ever overhear a conversation in a foreign language? You might pick up a word here and there, or make general observations about tone, emotion, and behavior. But the vast majority of what’s said is lost. This is situation most companies find themselves in when they get access to user data through Facebook Connect. There’s just so much being said, in a language that’s so difficult to understand, that most often very few if any actionable conclusions are drawn.

GraphDive is a social analytics and interest graphing company that enables enterprises to interpret this rich data stream. The company, which is focusing initially on the e-commerce, media and travel verticals, delivers personalized user recommendations through ranked user interest graphs and inferred demographic data including age, education, and marital status.

Founded in 2011, GraphDive opened its API to the public on an invite only basis approximately one week ago and announced $1 million in funding at the time from Crosslink Capital, Correlation Ventures, and several angels investors including early dropbox investor Pejman Nozad. Few outside the company’s limited customer list have gotten a look at the platform, so I was eager to take it for a spin this week.

Over its nearly two years in development, GraphDive has built an extremely sophisticated semantic analysis and machine learning system to understand the meaning and context of online social behavior. The results of this are quickly apparent when looking at the demographic and interest graph results delivered on an average user.

The company boils down each piece of social data, including comments, shares, likes, links, and even the age and actions of one’s connections to infer demographic insights about users. Most powerfully, the company delivers an estimated age range for every user, including the 80 percent who don’t list this data on their social networking profiles and others who list it incorrectly. According to co-founder and CEO Shahram Seyedin-Noor this inferred age has proven more than 90 percent accurate in early testing. Similar analysis is currently being conducted on education status and marital status inferences being delivered, but the already apparent conclusion is that GraphDive gives businesses a far clearer picture of individual users than is available otherwise.

This type of socially inferred demographic data is not available elsewhere, and accordingly sets GraphDive apart from the numerous other interest graphing companies. An accompanying dashboard provides companies with overall audience analytics, including details like the number of Facebook connected users on their platform and their average age, education, marital status (both actual and inferred).

On the interest graph side, the system is smart enough to understand and combine disparate terms like San Francisco, the Bay, the Mission, BART, and the Giants, wherever and however they appear, to deliver a single interest graph score for the topic, San Francisco Bay Area. The company’s knowledge graph knows that each of these items are connected and they each sit a specified distance apart from one another based on their interrelatedness. Based on this data, a final ranking is delivered for each topic, with underlying data-points and explanations available to the GraphDive user.

When Seyedin-Noor and I used his profile as an example, San Francisco was the top ranked interest on his personal interest graph, despite not being listed as his home or current city on his profile. Others, such as tennis and technology, were similarly culled from his profile data and ranked accordingly.

GraphDive uses these findings to serve personalized recommendations. Each user interest graph is quickly sortable into 30 ranked canonical categories, such as sports, culture, arts, or geography, against which companies can compare their content or product offerings to find relevant matches.

One potential shortfall of the product, is that it does not yet provide directional sentiment analysis, or in other words it can only tell if a user talks about Lebron James alot, not whether he loves or hates him. GraphDive’s founders will argue that what’s most important is that the user is passionate about the topic, but in plenty of instances this distinction would be valuable to businesses. Expect this ability to be added in the future.

GraphDive wraps all this functionality into two simple API, one for demographic inferences and one for interest graphs, whose ease of use belie their underlying power. Seyedin-Noor points to Klout as the role model in this regard and proudly says that even non-programmers can implement and extract value from its product. GraphDive is currently serving more than 1,000 API calls per day to its select early customers and found through a recent customer survey that 70 percent of GraphDive’s personalized results are “very relevant” to users.

Menlo Park-based GraphDive has less than ten full time employees and is in the process of scaling its team and infrastructure to meet the needs of a larger number of customers. Its data analytics engine works with any unstructured dataset, not just Facebook, and the company plans to extend its offering to other social platforms in the near future.

The company is still analyzing early customer activity to finalize its pricing model, although it says there may be multiple options based either on per call pricing, where industry averages suggest $0.01 to $0.75 per call, or a volume tiered monthly subscriptions. For clients that want customized “last mile implementations,” these are also available at premium pricing.

GraphDive has some competition in the interest graphing space, including well funded and well estabilshed publisher favorite Gravity. What sets it apart is that it’s the first to offer demographic inferences and offers a simplicity of API implementation that is extremely attractive.

Businesses will continue to get more and more data from online users. Those who draw the most accurate and actionable insights will be at a distinct advantage. GraphDive has rolled out an impressive first product and is likely to have a hand in determining these winners and losers.