摘要： 构造了多层Count-Min概要数据结构来概括流数据中的层次结构。通过定义多层数据域U*上两两相互独立的异或哈希函数族，将数据流元组映射到L×D×W的三维计数数组，L是层次个数，D是从哈希函数族中均匀随机选取的哈希函数个数，W是哈希函数的值域。基于该结构，利用广度优先查询策略，查找多层频繁项集和估计多层频繁项值。实验表明，该结构在更新时间、存储空间和估计精度方面比直接堆叠多个Count-Min结构有较大的提高。

关键词: 数据流, 概要数据结构, 频繁项集, 随机算法, 多层结构

Abstract: A hierarchical Count-Min (HCM) sketch data structure is implemented to summarize the hierarchical structure in stream data. By designing and defining a XOR-based pair-wise independent family of Hash functions on the hierarchical domain U*, the sketch maps stream data items to a three dimensional array of counters of size L×D×W. Of the counter array, L is the number of layers in hierarchy, D is the number of uniformly and randomly chosen Hash functions, and W is the range of the Hash functions. The HCM sketch uses breadth-first search strategy to identify and evaluate the hierarchical frequent itemsets. The experiments demonstrate the improvement in update time, space usage and evaluating accuracy over the straightforward CM-based approach.

Key words: data stream, sketch data structure, frequent itemset, random algorithm, hierarchy

中图分类号: 

• TP311