This paper presents a new method of interactive refinement of a knowledge base. The first step of our method is a validation stage which checks the consistency and the completeness of the knowledge base. Then, the second step is a refinement stage which aims to eliminate the anomalies (inconsistencies or incompleteness) that have been detected. The refinement stage is an iterative process performed in close interaction with the expert; the main step of this process consists of submitting to the expert a rule that must be repaired with a list of possible modifications for this rule. Our originality is that the expert has to take decisions about relevant modifications of a rule, which is an easy task for him because the rule represents an element of knowledge that he understands well. After each decision of the expert, our method updates the set of anomalies in order to apply criteria to select the following rule that needs to be repaired.
展开▼
