Towards a framework for large itemsets generation in association rules mining

The problem of computing all association rules, for prescribed minimum support and confidence, by analyzing a large data base of transactions, can be decomposed into two large building blocks; namely to find all the large itemsets, and to use these to generate the rules. In this paper we present a, framework for the former expressed as a generic algorithm that operates upon function objects and containers with their associated iterators. We show how the framework can be instantiated to produce implementations of Apriori, and Partition; two well-known approaches to large itemsets generation. We also show performance curves associated with implementations obtained by instantiating the framework for Apriori, using two different, backing data structures, namely hash trees and skip lists; the latter being a novel implementation on its own. Our framework has been implemented in C++ by using generic programming technology popularized by the Standard Template Library (STL).