Abstract
Many knowledge discovery tools have been developed using data mining, the integrating technologies of machine learging, database, statistics and others. We have been constructing "mondou" search systems based on extended association rules. In this paper, we discusse the experimental results of text mining applied for web hyper texts, INSPEC database, and magazines and articles index data in the National Diet Library. First of all, we express about the efficient strategies in order to derive association rules. Next, we discuss the relation between the threshold values and association rules, and we focus on the techniques of ROC (Reciever Operating Characteristics) graph to evaluate the characteristics of derived rules. By using the ROC convex full method, we can estimate appropriate threshold values to derive association rules for keywords.
