Semantic database modeling: survey, applications, and research issues

Reviewer: Martin E. Modell

Semantic data models represent an attempt to describe objects and the relationships between them from the viewpoint of their real-world existence, rather than from the way the data used to describe these objects are stored in the database. While many of these objects are physical and relatively easy to describe, many others are abstract, conceptual, and difficult to describe. Semantic data models provide mechanisms for dealing with and modeling data abstractions and “provide a higher level of abstraction for modeling data, allowing database designers to think of data in ways that correlate more directly to how data arise in the world.” The record-oriented data model world primarily uses three model constructs: hierarchic, network, and relational. These constructs traditionally revolve around physical objects (documents, people, places, and things) and cannot easily represent the more conceptual aspects of the real world. Semantic data models can capture concepts as well as data and are used in an attempt to describe how these physical objects relate to each other and how they are grouped (both physically and logically). The developers of the various semantic data modeling methods attempted to provide the means for translating the descriptions of the real-world objects into graphic representations to facilitate analysis and to assist the design process. Because these models provide more constructs and abstraction capabilities, they can represent a wider variety of situations and conditions. More than a dozen different modeling methods have been proposed over the years. Unfortunately, each technique uses a different set of terminology, a different set of constructs, and a different set of rules. This paper surveys the most prominent among these methods: the entity-relationship model, the semantic data model, and the functional data model. As a basis for discussing the various differences and similarities between the other models, the authors use a generic semantic model that “was developed expressly for this survey and is based largely on [the above three] most prominent models . . . along with many concepts derived from the IFO model.” The authors have a large topic to cover and many concepts to present. Their paper is lengthy (55 pages) and is supplemented by five pages of references. The paper has four main parts: philosophical considerations of semantic database modeling; a tutorial on semantic data modeling; a survey of prominent semantic data modeling techniques; and a discussion of the research, implementations, and theoretical analyses being performed in the semantic data modeling arena. The paper has much to commend it: the coverage of the various modeling techniques is complete and unbiased; the explanations of the concepts which underlie the various techniques are concise; and it is well illustrated and logically organized in general. The problem is that the paper is very difficult to read. This difficuty is due as much to the authors' writing style as to their method of presentation, the scope of what they try to cover, and the territory that they do cover. The large number of terms and concepts that are involved in so many different methods becomes confusing and difficult to follow after a while. Surveys and tutorials should be easy to comprehend so the reader can devote his or her energies to understanding rather then deciphering. By attempting to combine a survey and a tutorial in a single paper, the authors have done a disservice to their readers. Although the paper is informative, the authors would have done well to separate it into two parts—a survey and a tutorial. This would have allowed a much clearer and more extended discussion of the concepts. My final difficulty with this paper concerns the particular versions of the selected models that the authors use for discussions. Although many of these models were developed in the mid- to late 1970s, they have been extended and modified over the years in many significant ways. It appears that the authors have chosen to use the original developers' presentations as a basis of discussion rather than the current usages. Chen's entity-relationship model [1], for instance, has been extended and modified since 1976 by innumerable papers, conferences, and practitioners to the point where it is substantially different from the original proposal. Many of the other models have evolved in a similar manner. Obviously it is much more difficult to discuss the present state of these modeling techniques' evolution than it is to discuss the original versions. The current implementations of these models have come closer to each other than the paper implies. A follow-up paper might be in order.