User adaptive inference in expert systems

Abstract

Dialogues generated by the backward chaining control strategy is not adaptive to the user level. To overcome such a limitation, we propose the Question & Adapt (Q&A) approach. The Q&A approach generates dialogue at the user level regardless the depth of knowledge base. An issue is how to set and/or adjust user level so as to minimize the expected number of questions. We have developed several theorems and estimators about the average, variance and randomness of user levels. Such findings are applied to the design of user adaptive expert systems.