Entity Connectivity vs. Hierarchical Levelling as a Basis for Data Model Clustering: An Experimental Analysis

Abstract

AbstractData model clustering is the process of dividing large and complex models into parts of manageable size, in order to improve understanding and simplify documentation and maintenance. Based on theories of human cognition, a previous paper proposed connectivity (defined as the number of relationships an entity participates in) as a basis for clustering data models. This paper describes a series of laboratory experiments which evaluate the validity of this metric as a basis for clustering compared to hierarchical levelling, which has been the predominant approach used in previous research. The first two experiments investigate the relationship between the metrics and perceptions of importance, while the third experiment investigates their relationship to how people intuitively cluster entities. The results show that connectivity provides an empirically valid basis for clustering data models, which closely matches human perceptions of importance and “chunking" behaviour. No significant results were found for hierarchical level in any of the experiments. The high levels of statistical significance and effect size of the results for connectivity, together with their consistency across different domains and sample populations, suggests the possible discovery of a natural “law" governing data models.