Pattern Discovery in Conceptual Models Using Frequent Itemset Mining

Abstract

Patterns are recurrent structures that provide key insights for Conceptual Modeling. Typically, patterns emerge from the repeated modeling practice in a given field. However, their discovery, if performed manually, is a slow and highly laborious task and, hence, it usually takes years for pattern catalogs to emerge in new domains. For this reason, the field would greatly benefit from the creation of automated datadriven techniques for the empirical discovery of patterns. In this paper, we propose a highly automated interactive approach for the discovery of patterns from conceptual model catalogs. The approach combines graph manipulation and Frequent Itemset Mining techniques. We also advance a computational tool implementing our proposal, which is then validated in an experiment with a dataset of 105 UML models.