Let's continue to examine some of the problems associated with keyword-based ad targeting systems:

2) Ad Impressions are Wasted on Poor Targets
Non-optimal or “false” targets are people who are associated with keywords that do not accurately reflect who they are.  It is important for an ad-targeting system to be able recognize the words that are truly associated with an individual and to filter out those that do not represent a proper understanding of a user’s interests.

Would you send an ad for Britney Spears Lunch Boxes to someone whose profile looks like this:

“Metallica all the waaaaay, man! Megadeath still rocks, and yeah, I’m into Cannibal Corpse, Morbid Angel, The Exhumed, Obituary, Thine Eyes Bleed, Gorguts and we mustn’t forget Britney Spears!!!”

Hmmm… this would probably not be a good target, but most existing keyword-based ad targeting systems would shoot the Britney Spears ad to this page.  We should be concerned with marketing to the customer at the other end as a whole human being, not just a series of keywords.  So exploring the relationships between the various interests and ideas found in UGC is of paramount importance.

An observant human is able to know which words are out of place and therefore recognizes that some of the words are used as a joke or sarcastically but a machine might not be able to understand these kinds of nuances.  However, by understanding how interests hang together, as companies such as Peerset do, an automated system is able to discern what does and does not belong to the group of interests and concepts that it finds on a page of UGC.  In this way, the best possible targets can be identified and inappropriate ones can be filtered out.