基于GPU的闭合频繁项集挖掘方法

Closed Frequent Itemset Mining Method Based on Graphic Processing Unit

提出一种采用图形处理器挖掘闭合频繁项集的方法,用二进制数据表示项集,利用单指令多数据的体系结构实现并行计算,结合项集索引树,可以提高项集支持度计算和项集查找的速度。在2种数据集上的实验结果表明,该方法能够用更少的空间保存频繁项集的全部信息,并减少挖掘时间。

This paper considers a problem that to the best of the knowledge has not been addressed, namely, how to use Graphic Processing Unit（GPU） for mining closed frequent itemsets. The method employs a single-instruction-multiple-data architecture to accelerate the mining speed using a bitmap data representation of frequent itemsets, a further memory-based index tree is used to make search faster. Experimental results show that the algorithm can store all the information of frequent itemsets using less space, and achieve better performance in running time.