The Andes project routinely generates Bnets from a program. The program is
expressed as a rule-based system. It is interpreted with a forward chainer
(CLIPS) and a dependency graph is saved. The graph is converted to a Bnet.
The program is several hundred rules, and the networks can be as large as a
thousand nodes, depending on the problem that the program is solving.
Andes is a tutoring system, and the rule-based program is a model of how to
solve physics problems. It contains some buggy rules which represent common
student misconceptions about physics. When the forward chainer forms the
deductive closure of the rules, a dependency network for all "possible"
solutions is generated, including many incorrect solutions. We can then
diagnose what knowledge the student has by clamping the nodes that
correspond to vectors and equations entered by the student when solving the
problem. This is one form of probabilistic student modeling. Others are
reviewed in Jameson, A. (1995). Numerical uncertainty management in user and
student modeling: An overview of systems and issues. User Modeling and
User-Adapted Interaction, 5. For information on our technique, you can
download papers from http://www.pitt.edu/~vanlehn/andes.html.
Student modeling techniques such as ours can probably be applied to
troubleshooting of electromechanical devices and other types of model-based
diagnosis.
-- Kurt
Kurt VanLehn
Director, CIRCLE: Center for Interdisciplinary Research on Constructive
Learning Environments
Professor of Computer Science
Senior Scientist, Learning Research and Development Center
Senior Editor, Cognitive Science
University of Pittsburgh, Pittsburgh, PA 15260
voice: (412)624-7458 fax: (412)624-9149
email: vanlehn@pop.pitt.edu http://www.cs.pitt.edu/~vanlehn