Correctness principles for rule-based expert systems

摘要

This paper defines a set of acceptability principles for a rulebase. The principles go beyond mathematical correctness concerns to distribution and simplicity conditions that can signal the existence of errors or awkwardness in the rules. The principles are Consistency, Completeness, Irredundancy, Connectivity, and Distribution. The intent of these principles is to assist the rulebase designer in constructing a rulebase and validating its behavior.The five principles are implemented by mathematical and computational criteria that specify algorithms for analyzing rulebases. The Consistency criteria address the logical consistency of the rules, and can rightly be considered “correctness” criteria. The Completeness and Irredundancy criteria preclude oversights in specifications and redundancy in the rules, and are more like “reasonability” criteria for the terms in the rules. The Connectivity criteria concern the inference system defined by the rules, and are like completeness and irredundancy criteria for the inference system. Finally, the Distribution criteria are “esthetic” criteria for the simplicity of the rules and the distinctions they cause, as well as the distribution of the rules and the values implied by them.These procedures do not solve the (hard) problem of choosing a representation for the important features of the system being modeled, and turning the characteristics of the features into rules. They only allow a set of rules to be checked for the various criteria, so that many commonly occurring specification errors can be caught quickly. This paper discusses the formation of rulebases from a set of rules, not the formulation of rules from a system under study.