A Software Agent for Adaptive Visualization of Hyperspace Based on Clustering of Navigation Graphs

摘要

A software agent will be described that makes it possible to provide the visitor of an internet site with adaptive maps of the hyperlink structure. The maps represent estimations of important navigation decisions for the user depending on his own former partial navigation path and navigation information provided by other users. A user's navigation behaviour is characterized by the set of his navigation decisions in the hyperlink structure (navigation graphs). The problem of finding a suitable characterization of the distribution of these navigation graphs and the development of an estimation procedure for this distribution will be described. This estimation can be reached by defining different metrices or distance functions between graphs like string-edit- or largest-common-subgraph distance and by application of nearest neighbourhood clustering. The knowledge concerning the distribution is then used for a classification of a user and thereby an adaptive hyperspace view presentation using graph matching techniques. A visualization example will be presented.