Data modeling in UML and ORM revisited

Although the traditional entity relationship approach is still the most widely applied technique for modeling database applications, object-oriented approaches and fact-oriented approaches are being increasingly used for data modeling in general. The most popular exemplars of the latter two approaches are respectively the Unified Modeling Language (UML) and Object-Role Modeling (ORM). An initial, comparative evaluation of these approaches indicated that UML has benefits for object-oriented code design (e.g. implementation detail, including behavior), while ORM has advantages for conceptual data modeling (e.g. semantic stability, graphical expressibility; clarity and validation mechanisms). This paper further examines the relative strengths and weaknesses of ORM and UML for data modeling, focusing on attribute multiplicity, association arity, advanced constraints and subtyping. This analysis is given wider generality by addressing various language design principles (e.g. parsimony, orthogonality, convenience, expressibility) and illustrating how metamodel extensibility can be used to capture some features of one approach within the other.