This project explores website-based agents, designed to provide direction and recommendations to visiting clients. Time, processing power and intimate knowledge of the content being served are leveraged on the server-side, to allow suggestions and linkages to be pushed to users, based on their current interests and browsing history.
Agents that can accurately model their user's preferences, and communicate that information to one another as necessary, seem to be a focus of much recent industry attention. However, live users aren't the only information-producing bodies on the net; in fact, user-to-user interactions may make up just a small portion of useful net or Web traffic today. The bulk of information, from a consumer's point of view, sits on static web-sites, waiting to be explicitly pulled to a client. Site designers can only convey as much organizational information as fits on a navigation bar, and can't emphasize or de-emphasize content based on a user's interests. The information on a site is assembled according to how the site-designer thinks a user will want to browse -- and, while intelligent layout and authoring has a lot to be said for it, all hierarchies cannot be all things to all users.
Maitre-D is a preliminary step in addressing that issue. Using a combination of site-indexing, context-sensitive searches, and a layout/display system, it tries to convey "just-in-time" linkages between individual pieces of content. Rather than relying solely on fixed hyperlinks, it suggesting current contextually relevant locations the user may wish to experience.
Get the paper in HTML or PDF or PS form.