REASONING ABOUT PREFERENCE MODELS

Programs that make decisions need mechanisms for representing and reasoning about the desirability of the possible consequences of their choices. This work is an exploration of preference models based on utility theory. The framework presented is distinguished by a qualitative view of preferences and a knowledge-based approach to the application of utility theory. The design for a comprehensive preference modeler is implemented in part by the U tility R easoning P ackage (URP), a collection of facilities for constructing and analyzing preference models. Qualitative mathematical reasoning techniques are employed to develop partial specifications of single-attribute utility functions from qualitative preference assertions. Functions are described in terms of gross behaviors, symbolic forms, and parametric constraints. Appropriate dominance-testing algorithms are chosen from a knowledge base of stochastic dominance routines based on qualitative properties of the utility function. URP constructs multi-attribute utility functions from a set of independence conditions by applying proof rules from a knowledge base containing the important decomposition theorems from the literature. Proof rules describe the logical relations among independence conditions and functional forms. Hierarchical decompositions are structured automatically. Flexible model construction provides the potential for interpreting preference choices under a procedure that does not depend on the underlying utility model. Model-independent interpretation enables assessment under a wide range of descriptive theories of preference choice. Domain-specific preference knowledge is incorporated in URP by tying domain concepts to the modeler''s technical vocabulary. Development of a health preference knowledge base illustrates how preference modeling could be included in a knowledge-based system for a particular application.