This thesis proposes two schemes which can improve the efficiency of inference in rule-based expert system by reflecting the characteristics of users.
The first is the consideration of user``s level of expertise. The second is the reuse of dialog made in the past. To reflect user``s level of expertise, Question and Down approach is proposed. For the Question and Down method, the determination of down-size is very important. Therefore performance for various down size is evaluated. The implication for the design of expert system is derived from the performance evaluation. As a second type of support for user dependent dialog, the Dialogued Fact Base is proposed to reflect user``s dialog history.
Finally, a prototype UDDG(User Dependent Dialog Generator) which reflects the user``s characteristics discussed above is developed using pascal in the micro computer.