Improved Counter Based Algorithms for Frequent Pairs Mining in Transactional Data Streams

Abstract

A straightforward approach to frequent pairs mining in transactional streams is to generate all pairs occurring in transactions and apply a frequent items mining algorithm to the resulting stream. The well-known counter based algorithms Frequent and Space-Saving are known to achieve a very good approximation when the frequencies of the items in the stream adhere to a skewed distribution. Motivated by observations on real datasets, we present a general technique for applying Frequent and Space-Saving to transactional data streams for the case when the transactions considerably vary in their lengths. Despite of its simplicity, we show through extensive experiments that our approach is considerably more efficient and precise than the naïve application of Frequent and Space-Saving.