A constraint programming approach for quantitative frequent pattern mining

Abstract

Itemset mining is the first pattern mining problem studied in the literature. Most of the itemset mining studies have considered only Boolean datasets, where each transaction can contain or not items. In practical applications, items appear in some transactions with some quantities. In this paper, we propose an extension of the current efficient constraint programming approach for itemset mining, to take into account quantitative items in order to find patterns with their quantities directly on the original quantitative dataset. The contribution is two folds. Firstly, we facilitate the modelling task of mining problems through a new constraint. Secondly, we propose a new filtering algorithm to handle the frequency and closeness constraints. Experiments performed on standard benchmark datasets with numerous mining constraints show that our approach enables to find more informative quantitative patterns, which are better in running time than quantitative approaches based on classical Boolean patterns.