An Expert System Shell for Uncertain Rule- and Model-Based Reasoning

Detailed design assumptions and functional principles of two hybrid rule-and model-based expert system shells for uncertain backward and forward reasoning are presented. The systems may be used to reason with any knowledge base, which consists of following parts: • Uncertain Rule Base, containing uncertain IF...THEN... rules used for reasoning • Uncertain Model Base, containing exact arithmetic and uncertain relational models • Exact Constraint Base, containing sets of exact mutually exclusive and completely exhaustive askable conditions • Uncertain Constraint Base, containing sets of uncertain mutually exclusive and exhaustive askable conditions • Advice Base, being a directory of advice corresponding to rules of the rule base. • Advice Files, being text files with advice. The Knowledge Base semantics is simple, intuitive and straightforward. Practically unlimited nesting of rules, models as well as rules and models is allowed. Uncertainty is modelled using a modified Stanford Certainty Factor Algebra. The classic Stanford Certainty Factor Algebra using Certainty Factors in the range [-1,1] has been modified to make it conform to Aristotelian Logic for the case of CF’s being equal 1 or –1 and take into account dependent and independent lists of conditions. A unified string representation for conclusions, conditions, arguments, certainty factors and results contributes to the system flexibility and easy of programming. No attribute structure is defined for conditions of the rule base. Instead mutually exclusive attribute values are defined using an Exact Constraint Base and Uncertain Constraint Base. The systems are equipped with diagnostic facilities automatically checking the ruleand constraint bases for inconsistencies and redundancies, providing warnings and detailed diagnostic messages. A Reasoning Report and Diagnostic Report are produced on the fly. They may be saved for analysis and record keeping.