Raven: Bayesian Networks for Human-Computer Intelligent Interaction
Ole J. Mengshoel, David C. Wilkins · 2002
Bayesian Networks are a promising computational formalism for drawing conclusions from large amounts of intelligence analysis data. They are suited to this human-computer intelligent interaction task because they’re easily mapped onto a comprehensible graphical network representation; and because they are superb in domains with large amounts of uncertain data. This paper shows how the intelligence analysis task can be mapped into Bayesian networks. And it overviews two research contribution: two new heuristic algorithms for efficient inference and approximation of Bayesian networks; and algorithms to generate representative test sets for evaluation of faster methods of inference.