Mining Indirect Association Rules

Abstract

A large database, such as POS data, could give us many insights about customer behavior. Many techniques and measures have been proposed to extract interesting rules. As the study of Association rule mining has proceeded, the rules about items that are not bought together at the same transaction have been regarded as important. Although this concept, Negative Association rule mining, is quite useful, it is difficult for the user to analyze the interestingness of Negative Association rules because we would get them too many. To settle this issue, Indirect Association rule mining has proposed. In this paper, we propose a new framework of Indirect Association rule via a mediator and a new measure μ based on measures PA and PD due to Zhang to mine Negative Association rules effectively without the domain knowledge. The μ measure has the advantage over the IS measure that is proposed with the first framework of Indirect Association rule mining, and satisfies all of the well-known properties for a good measure. Finally, we are going to analyze the retail data and present interpretations for derived Indirect Association rules.