DAISY, an RER Model Based Interface for RDB to ILP

Abstract

In this paper, we propose an RER (Refined-Entity-Relationship) model, an extension of the ER (Entity-Relationship) model, which has an added feature of each entity attribute as well as each relationship being one indicating whether it can be derived from others or not. The purpose of an RER model is to apply ILP (Inductive Logic Programming), one of the most expressive machine-learning algorithms, to data mining by directly connecting RDB (relational databases) and ILP systems. We believe that the interface based on our model enables to easily make ILP systems access RDB in order to realize a powerful data mining system incorporating ILP. We examined the accuracy of the interface with benchmark data for evaluation of learning algorithms in artificial intelligence fields. Further, we tested our model on an email database to be supplied to an ILP system called PROGOL which in turn successfully derived a set of rules to assign each newly arrived email to the most suitable classification class based on its contents.