A BELIEF FRAMEWORK FOR ORDER-OF-MAGNITUDE REASONING AND OTHER QUALITATIVE RELATIONS

We present a new framework for the representation of qualitative knowl edge and in particular order-of-magnitude relations. The framework is based on the assignment of beliefs to intervals. Generic techniques ar e proposed for the manipulation of basic-probability densities and bel ief functions to make inferences. The framework allows a rigorous inte rpretation of qualitative knowledge without abandoning the informal ap peal of intuitive relations. Multiple order-of-magnitude relations can be accommodated and combined for the same pair of quantities. Relatio ns can be propagated to other pairs of quantities; the increasing unce rtainty is rigorously modeled, but overly conservative interpretations that use a single interval for each relation are avoided. (C) 1997 El sevier Science Ltd.