一种基于变尺度滑动窗口的数据流频繁集挖掘算法  

Variable slide window based frequent itemsets mining algorithm on large data streams

作　　者：朱小栋[1] 黄志球[1] 沈国华[1] 袁敏[1,2] 

机构地区：[1]南京航空航天大学信息科学与技术学院,南京210016 [2]湘南学院计算机科学系,湖南郴州423043

出　　处：《控制与决策》2009年第6期832-836,842,共6页Control and Decision

基　　金：航空科学基金项目(01I52058);湖南省自然科学基金项目(05JJ40102)

摘　　要：基于传统滑动窗口机制的数据流频繁集挖掘算法较多地考虑快速且精确的效果,而较少考虑数据流的时变特性.对传统的滑动窗口机制进行改进,同时考虑数据流的海量特性和时变特性,提出一种基于变尺度滑动窗口机制的数据流频繁集挖掘算法V-Stream.该算法采用事务链表组的概要数据结构,能够根据数据流的数据分布变化自适应调整窗口大小.Eclipse上的仿真实验结果表明,V-Stream相比Manku算法提高了挖掘数据流频繁集的时间与空间效率.Considering the problem that many traditional time window based data streams mining algorithms have good efficiency upon synthetic data streams but fail to be applied to real data streams in the fields such as celestial bodies movement and high energy particles outbreak, a variable slide window based frequent itemsets mining algorithm V-Stream upon data streams is proposed. The algorithm could self-adaptively adjust the size of time windows. In the algorithm, transaction list group is adopted as synopsis data structure. The experiments results in Eclipse indicate that the V-Stream algorithm is more effective than the Manku algorithm in term of temporal and spatial performance.

关 键 词：数据流 数据挖掘 滑动窗口 频繁集 关联规则 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论] TP18[自动化与计算机技术—计算机科学与技术]