A software agent for adaptive navigation support in a restricted internet area
This thesis deals with the development of a software system that helps a user to search for information in the World Wide Web. The particular problem considered here is support in a well-defined, restricted Web area. Two support strategies are considered. One strategy is to present a visitor views of a local hyperlink structure depending on the current position in hyper-space and previous navigation decisions. Main partial problems to realize such a support are dealt with, like the registration of user behavior, the registration of information about the Web area and the presentation of support information on the client side. In contrast to similar systems, the developed system may be applied by a large fraction of Internet users instantly. The only requirement on the client side is Java support by the browser. The second considered support strategy is an estimation of the relevance of data objects and sequences in the Web for a specific client. This estimation is based on the client's previous navigation behavior and registered navigation behavior of other users (collaborative filtering). The approach to estimate relevant data objects in this thesis is to predict a user's future data requests. For this purpose the presented system stores user information on the server side. User behavior is modeled by graphs, consisting of nodes representing requested data objects and edges representing transitions. A new method is presented to predict future navigation steps that is based on a distribution estimation of registered graphs and a classification of a new (partial) navigation profile with regard to the estimated distribution. The different steps of the presented algorithm are evaluated using generated and observed profiles.