Sub-graph inference of large scale rule-based system

Abstract

In the large scale rule-based systems, the search efficiency for inference becomes more crucial as the number of rules increases. This thesis therefore proposes a method of selective inference, named sub-graph inference, to enhance the search efficiency under the backward chaining inference environment. The key concept of sub-graph inference is that only the relevant rules for a certain goal are invoked during the inference so that the large number of irrelevant rules would not deteriorate the search efficiency. However, to make the sub-graph inference possible, the goal structure consistent with knowledge base should beforehand be organized. This thesis therefore has developed a mechanism to automate this goal structuring process. A prototype named SUGAR(Sub-Graph Automatic Reasoner) is developed on the microcomputer to illustrate our approach.