On Exploring the Power-Law Relationship in the Itemset Support Distribution

Abstract

Abstract. We identify and explore in this paper an important phenomenon that the power-law relationship appears in the distribution of itemset supports. Characterizing such a relationship will benefit many applications such as providing the direction of tuning the performance of the frequent-itemset mining. Nevertheless, due to the explosive number of itemsets, it will be prohibitively expensive to retrieve characteristics of the power-law relationship in the distribution of itemset supports. As such, we also propose in this paper a valid and cost-effective algorithm, called algorithm PPL, to extract characteristics of the distribution without the need of discovering all itemsets in advance. Experimental results demonstrate that algorithm PPL is able to efficiently extract the characteristics of the power-law relationship with high accuracy. In addition, while applying algorithm PPL prior to discover approximate frequent itemsets, a desired result can be effectively obtained, showing its prominent advantage of being an important pre-processing means for mining applications.

Exploring the Influence of Sampling on Pattern Support Distribution

“…Two recent papers [3,4] used empirical studies on various retail transaction datasets to demonstrate that power-law relationships exist in the pattern support distributions of real datasets. It is suggested that this is caused by self-similarity within the pattern support distribution.…”