Kurt Kammerer
Living Systems AG, Villingen, Germany.

___________________________

Web Personalization with living agents

The problem: Large Websites have several 1000s of visitors per day without
being able to tell one visitor from another. Visitors easily get lost in
thousands of webpages, as non-individual navigation and retrieval options
can not deliver the kind of personalized information/content, which a user
expects to get. In the real world, one-to-one marketing is a preprequisite
for their business. Consequently, they want one-to-one strategies to be
also applied to their virtual marketplace, i.e. their website. Visitors
with similar profiles could be addressed as a community.

Our living agents approach: We distinguish between content and user
profiles, each being presented by living agents. Profile metrics measure
the distance between profiles. Short distances signal some relevance,
whereas long distances signal low relevance. Example I: If the profile of
content A is adjacent to the profile of user B, there is a good chance,
that user A is interested in information B. Example II: If the distance
between profile of user B and profile of user C is short, they may belong
to the same user cluster (community). These profile considerations allow us
to present content to a user in his/her individual preference order. So,
content is presented in a benefit-oriented order with max. benefit content
being presented first. Profiles are dynamic and are updated with every step
a user takes. This tracking information feeds the user profile. Content
profiles are updated accordingly. So, the matching process between content
and users is an ongoing task and provides different results depending on
the very moment of matching (due to the dynamic nature of user and content
profiles.). Profile information can be used for both supporting one-to-one
strategies as well as the definition of communities (profiles within a
certain circle) in a profile datawarehouse.

Our cartoon-prototype: We have developed a prototype, which allows a user
to rate cartoons (political, business, daily events,....). The profile of
the user and the content are then updated accordingly. If the rating is
negative, the distance between those profiles increases and vice versa. The
dynamic nature of content and user profiles becomes obvious as the sort
order may change on user transactions.