Separator: Sifting Hierarchical Heavy Hitters accurately from data streams

Abstract

In this paper, we present a new algorithm, Separator, for accurate and efficient Hierarchical Heavy Hitter (HHH) detection, an emerging research area of data stream mining. Existing algorithms exploit either bottom-up or topdown processing strategy to solve this problem, whereas we propose a novel combination of these two strategies. Based on this strategy and a devised compact data structure, we implement our algorithm. It is theoretically proved to have tight error bound and small space usage. Comprehensive experiments conducted also verify its accuracy and efficiency. © Springer-Verlag Berlin Heidelberg 2007.