With the advent of integrated heterogeneous information sources, there is a renewed interest in information modeling (IM). The authors emphasize that most previous approaches to IM are based on the assumption of inherent classification, that is, that the first modeling step is usually to identify the classes or types of things needed to describe the domain. Their goal is to show that this assumption is inappropriate in many applications of IM. They propose a two-layered approach to IM. The first layer represents the existence of things with properties, independent of the classes to which the things belong. The second layer consists of class definitions based on sets of properties.
In section 2, some undesirable consequences of inherent classification are discussed. On the schema design level, the authors explain problems of multiple classification, view integration, schema evolution, and interoperability. Other problems arise in database operation. They include handling exceptional instances, reclassifying instances, adding and removing instances, and removing and redefining a class. Section 3 attempts to describe the theoretical foundations of classifications, with the help of ontology and epistemology. The authors use Bunge’s simple ontological model, which dates to the 1970s. They conclude that both ontology and classification theory recognize that instances exist independent of any classes.
The authors adapt these ideas in a two-layered approach in section 4. They draw two partial conclusions: that recognizing the existence of things should precede classifying them, and that there is no single “correct” set of classes to model a given domain of instances and properties. The particular choice of classes (a view) depends on the application. The two-layered approach consists of the instance model and the class model. In the former, instances and properties are represented. Instances can be queried and modified by primitive operations. Similar operations are designed for the class model.
In section 5 the authors present various implications of the two-layered model. They show that many problems connected to the assumption of inherent classification disappear in this approach. Section 6 discusses how to combine this approach with other models, particularly with the entity-relationship (ER) model and object models.
Section 7 addresses some issues related to the implementation, efficiency, and practicality of the model. The most interesting part concerns querying a database. The authors present queries that are not available in traditional class-based databases, such as relational databases. For example, we can obtain identifiers of all red cars stored in a database, regardless of the database columns in which they are placed. The chapter concludes with a detailed discussion of how to implement the two-layered approach.
The rest of the paper summarizes the main ideas of the model, points out its advantages, and gives a context for its application. Obviously, many other classical database issues can be reformulated and solved in this new paradigm.
The paper is well-arranged and sound. It offers an interesting approach to IM, which can make many contributions and has many consequences for the development of future database architectures.