Making important decisions often requires treating major uncertainty, long time horizons, and complex value issues. To deal with such problems, the discipline decision of analysis and knowledge-based systems have been developed at two different angles of decision system. However decision analysis has drawbacks such that the amount of effort expended and time spent on modeling a problem is too burdensome and that the resulting model is applicable to only one specific problem. This cost, in both time and money, has been a limiting factor in the use of decision analysis. Although many researchers are interested in developing methods for providing useful knowledge-based decision support, they have paid little attention to the modeling of user-specific preferences and tradeoffs about the quantitative values obtained from users. Furthermore, purely symbolic(non-quantitative) reasoning techniques have limited utility in the solution of many decision problems without the explicit consideration of the quantitative notions of uncertainty and tradeoffs.
Therefore, the aim of this research is to synthesize the methodologies of decision analysis with knowledge-based systems to utilize their benefits. To such a purpose, this dissertation suggested two integration directions:
1. Decision analysis is performed with knowledge-based systems-the knowledge of decision analyst(s) and domain expert(s) are converted into knowledge-based systems so it is possible to analyze a decision problem without the help of decision analyst and domain experts.
2. The realm of knowledge-based systems are expanded to cover the methodologies of decision analysis-the decision model is also a knowledge representation method so the methodologies of decision analysis may be compounded with the knowledge-based systems.
To replace the role of domain experts with knowledge-base, decision class analysis is formally defined. Analyzing a class of decisions makes it possible to build and analyze a decision model with a knowledge-based systems. Based on this concept, this research develops a value-preserving expansion procedures to build an influence diagram, as a decision model. And it is designed a knowledge-based decision system, KIDS which is an expert system to build an influence diagram. KIDS is implemented in Macintosh using Hypertalk and it suggests a decision for a raw-material buyer of CHEIL Synthetic Inc., which has produced textile products.
Next this research applies the decision-analytic techniques to existing knowledge-based systems. The methodologies suggested in this research are knowledge representation using decision tree, uncertainty handling using beta distribution, user-specific preference model, sensitivity analysis in a decision analysis cycle. The knowledge-based system utilizing these decision-analytic techniques are referred to decision-analytic consulting systems. The decision-analytic consulting system is superior to existing knowledge-based systems in respect of to treat explicitly the important uncertainties and tradeoffs that underlie symbolic reasoning. Despite of their benefits, it is difficult to explain the generated advice in a decision-analytic consulting systems. Thus, this research attempts to produce qualitative text explanation of treatment decisions based on explicit normative decision models. Especially, the sequential application of the five axioms for reasoning about preference makes it possible to explain the reason why a decision is generated. To implement these methodologies, a prototype decision -analytic consulting system is designed and implemented on the microcomputer using lisp. The operation of DACS is illustrated via a Cervical Cancer Treatment example.
The knowledge-based decision system or decision-analytic consulting system decrease the amount of time and effort for modeling and analyzing decision problems and, it is believed to increase the acceptance of decision analysis as a useful tool.
