Like KN-AHS, the HYPERFLEX system is an adaptive hypertext browser. However, in this case the adaptation is navigational, rather than presentational. One distinguishing feature of this system is that the user can always access all topics within the system because a full list of topics is always accessible. The adaptive component of the system orders the list such that topics considered most relevant to the current goal (if specified) or topic are placed at the head of the list.
The domain model underlying HYPERFLEX is a fully connected semantic network; that is, as we mentioned, every topic can be reached from each topic. This information is encoded as a topic to topic associated matrix. The values in this matrix indicate the strength of relationship between topics. In addition, a goal to topic associative matrix exists whose values indicate the relatedness of goals to topics. The value of the matrix weights are set by users’ interaction within the system and so represents the users’ view of the domain and forms the user model. The values or weights within the matrices are combined to determine the order of the suggestion list.
Relevance feedback is used to update the associative matrix weights. There are two modes of achieving this. In manual mode the user reorders the suggestion list, moving more relevant topics to the top. In automatic mode, the length of time spent looking at a topic is used as a guide to relevance. However as Kaplan himself acknowledges there are significant flaws with this strategy as there is no guarantee that the user is actually reading the topic for the full time that it is displayed.
Authored by Serengul Smith
E-mail to:
serengul1@mdx.ac.uk
School of Computing Science Middlesex University
Revised: September 1998