Mining the Local Dependency Itemset in a Products Network

Abstract

Many studies have been conducted on market basket analysis such as association rules and dependent patterns. These studies mainly focus on mining all significant patterns or patterns directly associated with a given item in a dataset. The problem that has not been addressed is how to mine patterns associated with a given item from the local view. This problem becomes very meaningful when the market basket dataset is huge. To address this problem, in this study, first, a new idea called “local dependency itemset” is put forward, which refers to patterns associated with the given item. Second, a framework of mining the local dependency itemset is presented. The framework has two steps, which are executed iteratively. One is expanding the local dependency itemset that initially consists of only the given item; the other is updating the local products network. Third, this framework is implemented by three different dependence indicators and a typical local community detection algorithm. The experimental results confirm that the local dependency itemset is meaningful.