Reports

Highlighting User Related Advice

McKeown, Kathleen; Weida, Robert Anthony

Research on explanation techniques for expert systems has demonstrated that (1) explanations are most effective when they address the user's needs and (2) it is necessary to augment explanations with information that is missing from the expert system‘s reasoning. It is our thesis that explanation content can also be improved by removing extraneous information from the system's reasoning and recognizing the remainder to emphasize user concerns. To test our ideas, we have developed an interactive natural language problem-solving system called ADVISOR which advises students on course selection. Previously, we have reported on our methodology for deriving user goals from the discourse, representing different points of view in the knowledge base and inferring user-oriented advice with a rule-based system that employs information from the appropriate perspective to address user goals. In this paper, we describe a model for pruning an explanation to highlight the role of the user's goal. The model is part of ADVISOR's natural language generation component. We demonstrate its efficacy with examples of different advice that ADVISOR provides for the same query in the context of different goals.

Subjects

Files

More About This Work

Academic Units
Computer Science
Publisher
Department of Computer Science, Columbia University
Series
Columbia University Computer Science Technical Reports, CUCS-345-88
Published Here
December 17, 2011