Contextual Advertising

Delivering more appropriate ads that convert to revenue

Growth of Online Advertising

There is a wall of money approaching online advertising as the internet medium has a population reach (>15%) which outstrips the budget allocations currently given by advertisers to online (approx 4%). Of course some advertisers are early adopters, but some of the traditional brands are going online - with the help of Search Engine Marketing agencies which assist with the bid purchase of search engine words. If a user goes to Google and sees text ads - an advertiser has previously bid to buy the ad opportunity - using the keywords you type in. This is advertising against the user's "statement of intent", which may be "cheap flights" or new "nano ipod".

Adsense Takes Over Adwords

But the bulk of Google' revenue has swung from Adwords ("statement of intent") to Adsense - which we could call "inferred statement of interest". Adsense places ads next to publisher content - helping the publisher to earn additional revenue to offset their editorial costs. Google crawls the publisher's site, establishes the best keywords, so that when a fresh user loads the particular page of content, Google can queue in the most relevant text ads. In theory it works well - except when the publisher changes the page of content without Google able to re-crawl the page. Also how does Google establish the best keywords? Not very well is often the verdict heard amongst publishers.

Most publishers sell their own advertising that goes at the portal doorway or front of the website. The home page gets the most traffic - so that is why the adverts earn them the most revenue. But what of the outlying pages that fewer people see: how can publishers optimize the revenue?

Traditional online publishing has revolved around display ads - so called banners and skyscrapers. A travel webpage gets a travel ad, based on the categories used by advertising networks like ValueClick, Blue Lithium and 24/7 Real Media to manually classify that page. But the user can be very confused. Why has a page about Alpine ski-ing got adverts about holiday sun in the Algarve?

Converting Categories to Keywords

Taking any publisher's webpage, Grapeshot's WordRank can establish the best ten words that describe the conceptual core of the document, and use those terms to request a keyword based ad. That means the publisher can now choose between a category ad paying $1 or a more specific keyword ad which may pay $1.50.

There is now an arbitrage opportunity to wrestle out the better paying ads whether they are category-based or keyword-based.

Mining Display Ads

Most advertisers are still clueless about the audience of people who see their graphic ads. Networks like Atlas and 24/7 can capture some intelligence about the user who clicks on an ad - the referring page, perhaps even a cookie ID. However with the URL of the page on which the ads occur, Grapeshot can extract the significant words and build a picture of that one user, using a cluster of top WordRank words inferred from what the user is reading and perhaps therefore interested in. Capturing one-off transactions is not as significant as building a growing and adaptive word profile based on successive transactions, behind each user, then queuing the next ad that one user should see - which could be a more personalized ad which has a more effective click through rate (CTR) and ultimate revenue (ROI).

Implicit Mining of the User's ClickStream

Whilst the pages people read are not explicit "statements of intent" in the same way a user looks for flights, ipods or cameras to actually buy via search engine; evidence does show that many people research a product, before they buy. It is therefore self-evident that if publishers can capture the content of the clickstream (which their logs already contain), then they can build up a profile of each and every unique user ID, by reference to pages seen. Grapeshot extracts the significant words for every document, using WordRank, and can alter the word weights for each user. That means amongst one million users with the word "ipod" in their profile - some people would have ipod at a +10 weight whilst others have it lower, say 8.2 or below. The philosophical leap is that Grapeshot can begin to permit advertisers to reach the top quartiles of an audience that likes a certain word - based on implicit analysis of the click stream. It means ads can be more contextually relevant to a user - such that punters get less ad noise, and advertisers get a stronger union of inferred intent with actual intent.

The Grapeshot SDK is a software toolkit designed specifically for OEM usage with a comprehensive API for delivering Intelligent Search Architecture (iSA) inside your own application. Built for a distributed, networked architecture, Grapeshot is small, robust and XML-based. It is ideally suited for embedding inside advanced advertisement network systems.

• Tell me more about Intelligence Search Architecture (iSA) and ADVERTiSA?