Inferring User-Oriented Advice in ADVISOR

Weida, Robert Anthony; McKeown, Kathleen

To be considered cooperative, an expert system must be easy to interact with. It must produce responses that are contextually appropriate, sensitive to the needs of users and suited to the user's level of sophistication. Expert system responses must be explained in a clear and concise fashion, often to untrained users. We have constructed a student advising system called ADVISOR, which contains a rule-based expert system, to test our ideas in natural language understanding, user modeling, use-oriented explanation, and text generation. ADVISOR assists computer science majors in course selection by providing information and offering advice during a natural language question-answering dialogue. In previous papers, we presented an overview of ADVISOR and detailed out methodologies for deriving user goals from the discourse and expressing expert system reasoning in natural language that emphasizes use goals (McKeown et al., 1985; McKeown, 1988; McKeown and Weida, 1988). In this paper we focus on how the expert with supporting justifications and additional relevant observations. Since advice is geared to the individual student, ADVISOR can provide different answers to the same questions as well as different explanations for the same answer. Indeed, its advice may change, along with the student‘s goals, as conversation progresses.



More About This Work

Academic Units
Computer Science
Department of Computer Science, Columbia University
Columbia University Computer Science Technical Reports, CUCS-346-88
Published Here
December 17, 2011