Inference Graphs: A Structural Model and Measures for Evaluating Knowledge-Based Systems

The initial objective of the research was to determine whether, and if so how, measurement-based techniques could contribute to the Quality Assurance of knowledge-based systems.
The investigation identified several quality factors of interest: usability, maintainability, and related attributes; reliability; correctness; and efficiency. Usability and some aspects of correctness were considered to be addressed by existing methods. For the remaining factors, the conclusion was that many if not all could be addressed by a graph-based, structural model of the knowledge base.
The Inference Graph model was developed to satisfy these requirements. A number of attributes of the Inference Graph were identified and the means to construct the graphs and to measure these attributes were formalised.
An experiment was constructed based on a simple CASE tool using Inference 7 Graphs. From the results of this experiment, it was seen that the tool and the underlying methodology could be used for several quality-related tasks: To classify rule-based systems; to evaluate or predict complexity and modularity; to identify and optimise resource-intensive areas of the rulebase; and to identify appropriate inference mechanisms. Quality Assurance based on these activities is expected to lead to improvements in maintainability, correctness and efficiency, thus satisfying the proposed requirements.
Proposals for future extensions to this work include the upgrading of the CASE tool to meet the needs of ongoing project development, and the addition of a dynamic element to the Inference Graph model.