STRATEGIES FOR EXPERT CONSULTATION IN THERAPY PLANNING (ARTIFICIAL INTELLIGENCE)

摘要

Many artificial intelligence researchers have recently focused on determining appropriate knoweldge representations for achieving high performance from knowledge--based systems. Production rule systems have become quite popular for problem solving in medicine as well as in other scientific domains. This work presents several new schemes that work within a general production rule representation to increase efficiency and expressive power.;Strategies for several aspects of the therapy planning problem are proposed. Therapy planning has been handled successfully for several applications in "real world" domains using a single system of generalized primitives. Treatment selection is handled via a new type of rule termed a precedence rule. A precedence rule consists of a topological sorting procedure within a general production rule representation. Precedence rules can be easily and succinctly represented and provide a qualitative approach to therapy selection and merging that we have found natural for our applications. Time-dependent variables are handled using a functional time windowing technique. This technqiue is used to select and process various model parameters. The addition of the time analysis capability required no modification to the production rule handling routines. Thus, the efficiency of the system in handling the rules was not degraded. The strategies used can be shown to be more efficient in both storage and time than previous schemes.;Artificial intelligence research has placed increasing emphasis on consultation systems that not only perform well, but can also provide explanations of their behavior. Precedence rules naturally provide such a capability. The explanations are more than just traces through the reasoning process. The precedence scheme provides the rationale behind each choice and comparisons between the competing choices. A precedence rule has the capability to summarize a great many facts elegantly.;The strategies for consultation modeling are the result of generalization over several large consultation systems. These strategies are currently being used in rule-based expert consultation systems for the diagnosis and treatment of ocular herpes, the diagnosis and treatment of rheumatic disorders, and isoenzyme interpretation for screening heart attack patients. Extensive examples from these systems are included.