Surprise-based qualitative probabilistic networks

Abstract

This paper discusses a modification of the kappa measure of surprise and uses it to build semi-qualitative probabilistic networks. The new measure is designed to enable the definition of partial-order relations on its conditional values and is hence used to define qualitative influences over the edges of the network, similarly to Qualitative Probabilistic Networks. The resulting networks combine the advantages of kappa calculus of robustness and ease of assessment and the efficiency of Qualitative Probabilistic Networks. The measure also enables a built-in tradeoff resolution mechanism for the proposed network. © 2009 Springer Berlin Heidelberg.