Dependency analysis for knowledge validation in rule-based expert systems

Keeping knowledge consistent is an important topic in the life cycle of developing expert systems. In this paper, we focus on some major problems in knowledge validation: redundancy, subsumption, cycles, conflict, and unnecessary conditions, and describe how these problems are solved in rule-based expert systems using dependency analysis. A rule-dependency graph is developed to describe the dependency relationship among the rules contained in a knowledge base. Since each type of inconsistent knowledge presents a specific topology in the rule-dependency graph, knowledge validation can be done by examining the structure of the graph. With the aid of the rule-dependency graph, we have developed a token-flow paradigm that identifies the inconsistent structure in the rule base. The idea is effective and can be easily implemented. Properties of our method are explored. Some practical examples are also presented