Fast algorithm for enumerating frequent itemset pairs in database of transactions

In this paper we present efficient algorithm for enumerating frequent pairs in database of transactions. Counting support for itemset pairs can be dominant step in frequent itemset mining occupying in some cases greater that 90% of total execution time. For efficient implementation we use one-dimensional array and define one-to-one function that maps 2-itemsets to array indexes preserving lexicographic order. In that way 2-itemsets are not generated explicitly; they are represented by indexes in one-dimensional array while their support counts are stored in appropriate array elements.