How to find frequent patterns?

An improved version of DF, the depth-first implementation of Apriori, is presented. Given a database of (e.g., supermarket) transactions, the DF algorithm builds a so-called trie that contains all frequent itemsets, i.e., all itemsets that are contained in at least `minsup' transactions with `minsup' a given threshold value. In the trie, there is a one-to-one correspondence between the paths and the frequent itemsets. The new version, called DF+, differs from DF in that its data structure representing the database is borrowed from the FP-growth algorithm. So it combines the compact FP-growth data structure with the efficient trie-building method in DF.