Meta-Rule-Guided Mining of Association Rules in Relational Databases

A metarule guided data mining approach is proposed and studied which applies metarules as a guidance at nding multiple-level association rules in large relational databases. A metarule is a rule template in the form of \P1 ^ ^ Pm ! Q1^ ^Qn", in which some of the predicates (and/or their variables) in the antecedent and/or consequent of the metarule could be instantiated. The rule template is used to describe what forms of rules are expected to be found from the database, and such a rule template is used as a guidance or constraint in the data mining process. Note that the predicate variables in a metarule can be instantiated against a database schema, whereas the variables or some high-level constants inside a predicate can be bound to multiple (but more speciic) levels of concepts in the corresponding conceptual hierarchies. The concrete rules at diierent concept levels are discovered by a progressive deepening data mining technique similar to that developed in our study of mining multiple-level association rules. Two algorithms are developed along this line and a performance study is conducted to compare their relative eeciencies. Our experimental and performance studies demonstrate that the method is powerful and eecient in data mining from large databases.