Automatic elicitation of interactive rules from data with exceptions using TMS

A method for eliciting interactive rules from data with exceptions is described. This method consists of the following three steps: create a hypothesis set (rule candidates); remove exceptional data; and choose the appropriate hypothesis. For the knowledge elicitation procedure, a TMS (truth maintenance system) is useful in choosing an appropriate hypothesis and detecting exceptional data candidates. The advantage in using TMS is that rules can be incrementally elicited from the data. The validity of this method is evaluated using a simple system which elicits rules about chemical reactions from a practical chemical reaction database. A comparison of results for this method and a statistical method shows that it is more useful in eliciting interactive rules.<>