Lest you think that this business of content and advertiser targeting is a simple matter of Internet coding, think again. At least one company is building its technology modeled on the most complex realms of the human genome itself. The company, NetSeer, says targeting can match human cell biology in terms of complexity and effective growth.
“When you look at the data of how genes regulate proteins you start to get a different sense of connections and groupings,” says John Mracek, recently-named CEO of NetSeer. “We have adapted research from the genome to develop mathematical tools that cluster pages into high connectivity groups that goes beyond the simple keyword search.”
The company has recently rolled out its first example of “concept-based advertising” called ContextLinks for Publishers. Its algorithms are based on recent, breakthrough techniques for understanding the metabolic processes in living cells, for answers. According to Mracek, NetSeer capitalizes on UCLA professor Vwani P. Roychowdhury’s work which has discovered a top-down analytical framework that could effectively partition hundreds – even thousands – of cell protein interactions into high-level, functional groups with absolutely no need to know what the groups actually did. In other words, past history of metabolic processes can feed an algorithm that can predict what that cell process will be. Roychowdhury’s work in 2005 connecting neural networks to spam filters was highly regarded, and adopted by many companies.
Here’s the connection for ad targeting. Serving up ads based upon a user’s keyword search entries has been successful because the user himself codifies his intent. But what do you do where there is no search? How can you monetize the white space on literally hundreds of millions of pages that are viewed each day, without knowing the conceptual framework that brought that user to that page at that moment?
“Using essentially the same top-down algorithms that succeeded in partitioning a cell’s processes into ‘islands’ of functionality, we have been able to examine the entire Internet – literally trillions of words – and identify islands of concepts – complete with the strength of their representation across the Internet – and the strength of their relationship to other concepts,” says Mracek.
The concept “bass” has a strong relationship, for example, to fish, fishing vacation, tuner, shoes, speakers, etc., while the concept “bass fishing” has no relationship to “stereo system” or “footwear sales extravaganza”). With more than 50 million concepts in its data set NetSeer devised a way to parse any Web page, in real time, into sets of those concepts. And since it works with concepts, rather than keywords, it says it can definitively match ad placements with the actual concepts that brought users to the page. So real-time, intuitive targeting can be approached.
Mracek says NetSeer is able “to consistently equal or outperform keyword-based ad placement with or without a keyword. NetSeer has seen a 2-to-4 times improvement in click through ratios over context-based approaches, and Mracek believes that concept-based targeting will do no less than change the “fundamental DNA” of Internet advertising. He also says the company is exploring licensing the technology to other vendors. Currently the company is not disclosing its client list, but Mracek says he expects key pilot results to be released next year.
