动态划分序列模式挖掘算法  

A Dynamic Partition Sequential Pattern Mining Algorithm

作　　者：夏岩[1] 倪世宏[1] 王彦鸿[1,2] 

机构地区：[1]空军工程大学工程学院,陕西西安710038 [2]西北工业大学自动化学院陕西西安710072,710072

出　　处：《计算机仿真》2009年第2期127-130,共4页Computer Simulation

摘　　要：在对海量数据进行序列模式挖掘时,由于产生的候选序列数量大并且需要反复遍历数据库,会带来存储受限及时效性问题。为此,基于无共享分布式环境,提出一种动态划分序列模式挖掘算法DPSPM。通过分布数据降低数据库扫描的规模,降低I/O开销,使用分派函数指定选举节点,降低通信开销,设计动态列表存储候选序列,降低内存开销,多个子过程异步运行,提高算法的执行效能。实验结果表明,DPSPM算法在较小支持度情况下对中大型数据库有较好的搜索效率。When mining sequential patterns among vast data, since huge amounts of candidate sequences are generated and the database is scanned for so many times, the memory is limited and the searching efficiency is low. In order to solve the problems, a dynamic partition sequential pattern mining algorithm （DPSPM） was put forward based on shared -nothing distributed environment. The I/O cost, communication cost and memory cost were reduced by data distributing, allotting function and dynamic list, the efficiency of algorithm is improved by asynchronous process. The experiment result demonstrates that DPSPM has high searching efficiency in the case of large database and low support threshold.

关 键 词：数据挖掘 序列模式 分布计算 动态列表 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]