CBP: A new efficient method for mining multilevel and generalized frequent itemsets

Abstract

The taxonomy(is-a hierarchy) data exists widely in retail, geography, biology and financial area, so mining the multilevel and generalized association rules is one of the most important research task in data mining. Unlike the traditional algorithm, which is based on Apriori method, we propose a new CBP (correlation based partition) based method, to mine the multilevel and generalized frequent itemsets. This method uses the item's correlation as measurement to partition the transaction database from top to bottom. It can shorten the time of mining multilevel and generalized frequent itemsets by reducing the scanning scope of the transaction database. The experiments on the real-life financial transaction database show that the CBP based algorithms outperform the well-known Apriori based algorithms. © 2008 Springer-Verlag Berlin Heidelberg.