Certainty Factor Model and Its Basis in Probability Theory

摘要

The certainty factor model proposed by Shortliffe and Buchanan (1984) for handling uncertainty in rule-based top-down reasoning expert systems is considered. The basic notions of rule-based top-down reasoning expert systems that are used in the model, such as production rules and derivations, are defined to allow a formal description of the model. In this formalization, the functions used in the model are extended with the notion of derivation. In the model, two basic measures of uncertainty are defined: the measures of belief and disbelief. These measures are defined in terms of a probability set function, and are approximated. It is shown that in some cases these functions respect the probabilistic basis by making rather strong assumptions. In other cases the approximation functions cannot be shown to be consistent with the foundation in probability theory suggested by Shortliffe and Buchanan. In actual implementations of the certainty factor model, the measures of belief and disbelief are not used. A third function derived from these two basic measures is used: the (redefined) certainty factor function. This function is defined in terms of the measures of belief and disbelief. For the purpose of subsequently computing certainty factors again an approximation function is defined. This approximation function respects the definition of the certainty factor function.