Finding frequent items in time decayed data streams

Abstract

Identifying frequently occurring items is a basic building block in many data stream applications. A great deal of work for efficiently identifying frequent items has been studied on the landmark and sliding window models. In this work, we revisit this problem on a new streaming model based on time decay, where the importance of every arrival item is decreased over the time. To address the importance changes over the time, we propose a new heap structure, named Quasiheap, which maintains the item order using a lazy update mechanism. Two approximation algorithms, Space Saving with Quasi-heap (SSQ) and Filtered Space Saving with Quasi-heap (FSSQ), are proposed to find the frequently occurring items based on the Quasi-heap structure. Extensive experiments demonstrate the superiority of proposed algorithms in terms of both efficiency (i.e., response time) and effectiveness (i.e., accuracy).