EzSeer, a spin-off company of the University of Alberta, has acquired an exclusive license of WebIC technology from Oct. 2007. If you want to leanr more about WebIC and EzSeer, please visit EzSeer Inc.

WebIC, is a client-side Web recommender system that predicts the user's information need based on his browsing patterns, then points him to webpages, from essentially anywhere on the Web, that contain information useful to that user.

To make this concrete: Imagine you are watching someone browse the web, and you see her examining the page of links, labeled "P1" above. She clicks on a link whose anchor includes the word "Dolphins", which sends her to a page (P2) about the Miami Dolphins, a football team in the NFL. She immediately clicks back to P1, and on that page, then skips the second entry with the anchor "free dolphin", and clicks on the third link, with anchor "Whale". She reads P3, then follows a link with the anchor-words "whale" and "Shamu", until she reaches P5, which involves "whales" and "seaworld".

At this point, you would probably assume the user is interested in "whale" and "dolphin", but not "football" or "NFL". This is because you saw the user follow links that include these first words, and back-out of a page that involves the second category. We can therefore characterize which words are "Information Content" (IC) words, in the context of the current session, based on its "browsing properties" -- e.g., how often the user followed an anchor that included that word, or backed out of a page whose title included that word, etc. We can then use these IC-words to identify which pages are likely to be useful --- "Information Content" (IC) pages --- as the pages that include many of these IC-words. And perhaps we could find the IC-pages by using the IC-words as the query terms in a search engine (like Google).

Our WebIC system is based on this idea. At run time, it collects the browsing properties of every word that appears in the user's current session, then uses a classifer to predict which of these words is likely to be an IC-word. It then has various ways of using the resulting set of IC-words to seek IC-pages.