The discovery of association rules can be divided
into two phases [AIS93]. First, discover all frequent
itemsets X E R. Then, for each frequent X, test for
all non-empty subsets Y C X if the rule X Y * Y
holds with sufficient confidence. The second part of the
problem can be solved in main memory in a straightforward
manner once the frequent sets and their frequencies
are known. Discovery of the frequent sets is
the hard part of the problem: with m items., there are
2”’ potentially frequent itemsets. For all but the smallest
m, efficient methods for locating the frequent sets
are needed. For the rest of this article, we consider the
task of discovering all frequent sets.