Hmine-rev:Toward H-mine Parallelization on Mining Frequent Patterns in Large Databases
Bowo Prasetyo, Iko Pramudiono, Masaru Kitsuregawa · IPSJ SIG Notes · 2005
Bowo Prasetyo Iko Pramudiono Masaru Kitsuregawa [email protected] [email protected] [email protected] University of Tokyo NTT Information Sharing Institute of Industrial Science, Platform Laboratories University of Tokyo NTT Corporation H-mine is a frequent pattern mining algorithm that takes advantage of a hyper-linked H-struct data structure, runs fast in memory-based setting, and is known to have high performance in a sparse data set. However, H-mine's inherent necessity to dynamically adjust H-struct links in the middle of mining process makes it difficult to do any parallelization effort on the algorithm. In this study, we propose a revised algorithm of H-mine that does not need any adjustment of H-struct links by modifying link structure and reversing the order of processing data. The revised algorithm has comparable performance with the original version and can be easily extended to use in parallel environment.