Since Chen’s seminal article [1], the entity-relationship (ER) conceptual model has been a major component of database design. Mappings exist for converting schemas from this model and its extensions (generally referred to as EER, for “enhanced” or “extended” ER) to relational and object implementations, inter alia. Its key ingredients are entities, relationships, and attributes, where relationships are associations among entities. Quite often, different choices may be made in capturing real-world data in this model. Since modelers typically find the notion of a relationship more difficult to appreciate than the notion of an entity, Wand et al. explore the semantics of relationships from an ontological point of view.
The background in ontology is provided in the spirit of Bunge [2]. The authors propose a structure of conceptual modeling based on ontological foundations. Table II is useful for comparing ontological constructs with commonly used modeling constructs (which may lead to errors and problems) and with their own proposed conceptual modeling constructs. Based on this correspondence, they enunciate a set of seven rules for modeling and then apply those rules to relationship types in ER modeling. The authors provide some helpful illustrative examples.
The problems I have with this paper stem from some of the conclusions, which represent a stretch from standard ER modeling. At times the authors contradict Chen’s model, such as when they forbid relationships from having attributes of their own. The definition of relationships as subsets of the Cartesian product of their entity sets plays little, if any, role in the discussion. Relationships that have a partial participation constraint on one of their entity types are discouraged (in favor of total participation on the appropriate subclass). Also, it is not clear that the rules lead to improved modeling performances. Batra et al. [3], cited by the authors, does admit that relationships are harder to understand than entities (which should not be surprising, since they are one level more abstract), but the main thrust of that paper is that EER modeling is superior to direct relational modeling.