Wei-Min Shen, Bing Leng
Metapatterns (also known as metaqueries) have been proposed as a new approach to integrated data mining, and applied to several real-world applications successfully. However, designing the right metapatterns for a given application still remains a difficulty task. In this paper, we present a metapattern generator that can automatically generate metapatterns from new databases. By integrating this generator with the existing metapattern-based discovery loop, our system has now become both interactive and automatic. It can suggest new metapatterns for humans to choose and test, or pursue these metapatterns on its own. This ability not only makes the process of data mining more efficient and productive, but also provide a new method for unsupervised learning of relational patterns. We have applied this method to several simple databases and obtained some encouraging results.