Presenting significant information in expert system explanation

Abstract

This paper presents a method for eliminating insignificant portions of an explanation of a conclusion — those portions that include terminology and inferences that the user does not have the expertise to understand, and those portions that add little to the user's belief in the conclusion. The method exploits a user model to select for presentation only those portions of an expert system's reasoning that add significantly to the user's belief that the conclusion is the right one. Examples demonstrate how the method generates concise explanations with only significant information, and how it tailors the explanation to the user.

This research was supported by an ERCIM Post-Doctoral Fellowship 94–05. Thanks to Cindy Wolverton for the signal processing example of Sec. 3, and to Myles Chippendale and Gerhard Wickler for helpful comments on an earlier draft of this paper.