Integrating Expectations from Different Sources to Help End Users Acquire Procedural Knowledge

Role-limiting approaches using explicit theories of problem-solving have been successful for acquiring knowledge from domain experts1. However most systems using this approach do not support acquiring procedural knowledge, only instance and type information. Approaches using interdependencies among different pieces of knowledge have been successful for acquiring procedural knowledge, but these approaches usually do not provide all the support that domain experts require. We show how the two approaches can be combined in such a way that each benefits from information provided by the other. We extend the role-limiting approach with a knowledge acquisition tool that dynamically generates questions for the user based on the problem solving method. This allows a more flexible interaction pattern. When users add knowledge, this tool generates expectations for the procedural knowledge that is to be added. When these procedures are refined, new expectations are created from interdependency models that in turn refine the information used by the system. The implemented KA tool provides broader support than previously implemented systems. Preliminary evaluations in a travel planning domain show that users who are not programmers can, with little training, specify executable procedural knowledge to customize an intelligent system.