Jeffrey Parsons

How UML is used

Many UML projects are not Use Case driven.

Theoretical foundations for conceptual modelling in information systems development

Conceptual modelling in information systems development is the creation of an enterprise model for the purpose of designing the information system. It is an important aspect of systems analysis. The value of a conceptual modelling language (CML) lies in its ability to capture the relevant knowledge about a domain. To determine which constructs should be included in a CML it would be beneficial to use some theoretical guidelines. However, this is usually not done. The purpose of this paper is to promote the idea that theories related to human knowledge can be used as foundations for conceptual modelling in systems development. We suggest the use of ontology, concept theory, and speech act theory. These approaches were chosen because: (1) they deal with important and different aspects relevant to conceptual modelling and (2) they have already been used in the context of systems analysis. For each approach we discuss: the rationale for its use, its principles, its application to conceptual modelling, and its limitations. We also demonstrate the concepts of the three approaches by analysing an example. The analysis also serves to show how each approach deals with different aspects of modelling.

The Medium Is Not the Message: Advertising Effectiveness and Content Evaluation in Print and on the Web

ABSTRACT Some have argued that traditional principles of mass media advertising do not apply on the web. We present an empirical study that contradicts this assertion. Our findings suggest that advertisers need not take full advantage of the enhanced capabilities of the medium to produce effective web advertising. Given equal opportunity for exposure to the target audience, the same advertisements were equally effective in print and on the web. However, for promotional material that consumers would not classify as advertising, evaluations were lower when the material was presented on the web. We propose a plausible explanation for this apparent paradox.

Emancipating instances from the tyranny of classes in information modeling

Database design commonly assumes, explicitly or implicitly, that instances must belong to classes. This can be termed the assumption of inherent classification. We argue that the extent and complexity of problems in schema integration, schema evolution, and interoperability are, to a large degree, consequences of inherent classification. Furthermore, we make the case that the assumption of inherent classification violates philosophical and cognitive guidelines on classification and is, therefore, inappropriate in view of the role of data modeling in representing knowledge about application domains. As an alternative, we propose a layered approach to modeling in which information about instances is separated from any particular classification. Two data modeling layers are proposed: (1) an instance model consisting of an instance base (i.e., information about instances and properties) and operations to populate, use, and maintain it; and (2) a class model consisting of a class base (i.e., information about classes defined in terms of properties) and operations to populate, use, and maintain it. The two-layered model provides class independence. This is analogous to the arguments of data independence offered by the relational model in comparison to hierarchical and network models. We show that a two-layered approach yields several advantages. In particular, schema integration is shown to be partially an artifact of inherent classification that can be greatly simplified in designing a database based on a layered model; schema evolution is supported without the complexity of operations currently required by class-based models; and the difficulties associated with interoperability among heterogeneous databases are reduced because there is no need to agree on the semantics of classes among independent databases. We conclude by considering the adequacy of a two-layered approach, outlining possible implementation strategies, and drawing attention to some practical considerations.

Choosing classes in conceptual modeling

ion from Instances. A class can be defined only if there are instances in the relevant universe possessing all properties defining the class. Adherence to this principle supports the fundamental meaning of concept—a specification of the properties common to some instances. If no instances from the relevant universe share a specified set of properties, there is no “sameness” and no economy of representation. It would be cognitively wasteful to store such a set of properties as a concept, as it tells us nothing about the relevant universe. This principle is perhaps obvious for knowledge acquisition methods that build up abstract descriptions from a given set of objects. However, it is less obvious for methods based on inferring new categories from existing ones [5, 12], as these methods do not use instances to identify candidate classes. Moreover, abstraction from instances does not appear as an explicit requirement in object-oriented design and semantic data modeling. COMMUNICATIONS OF THE ACM June 1997/Vol. 40, No. 6 65 Maximal Abstraction. A relevant property possessed by all instances of a class should be included in the class definition. This principle supports both cognitive economy and inference. If certain properties shared by all instances are not part of the class (concept) definition, some similarities among the instances in the class are lost. Moreover, properties shared by all instances, but not included in the classification, cannot be inferred if an instance is identified as a member of the class. Abstraction from instances and maximal abstraction are requirements for a class to effectively express what the instances in a set have in common. We therefore propose that these principles be used as necessary requirements for a set of properties to be considered a class. We consider a set of properties a potential class in a relevant universe if and only if: • It has a nonempty extension (at some point in time); and • It contains all properties common to all instances in the extension. Note that although the set of properties forming a potential class is fixed, the properties possessed by an instance may change with time. As an instance acquires or loses properties, it might “move” in and out of a potential class. In other words, the population of a class might vary with time. It follows from the definition that two distinct potential classes cannot be defined by the same set of properties. In practice, if distinct names have been attached to identical sets of properties, these names are synonyms. Also, two distinct potential classes cannot have the same extension, that is, at every time, there is at least one instance in one class that is not included in the other. The implication of the second observation can be illustrated by an example. Consider a university department having a policy that only graduate students can serve as research assistants and that all of them must do so. Although “graduate student” and “research assistant” may reasonably be thought of as distinct classes, the fact that they must always have the same instances means they cannot constitute distinct potential classes. Even though different properties could be chosen to define graduate student and research assistant, the same instances must share these properties. Maintaining the two as distinct classes is redundant. Moreover, it might lead to misunderstandings, since for many purposes people think of “graduate student” and “research assistant” as different concepts. For instance, a user in the campus health center may reasonably distinguish graduate student and research assistant classes, believing instances of each carry different forms of health insurance. This example shows how examining class extensions can help avoid misconceptions arising from prior informal “conceptual baggage.” The analysis in this example required that the extension of the classes be known. In general, to prevent redundancy, we propose that classification strategies check whether different classes have identical extensions. This does not imply a need for complete enumeration of instances, which, in most cases, may be impractical. Instead, it may be possible to identify synonymous classes by some reasoning. For example, a student organization may use the concept “member.” If all students of a university are members of this organization and only students are members, the concept “member” can be identified as synonymous with “student” without examining the actual instances. Class Structures The potential class concept restricts the possible classes formed to model the relevant universe. Whether a given collection of properties is a potential class can be determined irrespective of the other classes being considered. However, in conceptual modeling we are interested in collections of classes (or concepts). We now turn to the implications of applying cognitive economy and inference to such collections. Inference relates to the ability to infer properties from class membership. Cognitive economy involves balancing the objective of maximizing information content with that of minimizing the number of stored concepts, by emphasizing the value of distinguishing between concepts whose instances possess “meaningful differences.” Thus, the full information about an instance is usually distributed among all classes to which it belongs. To eliminate loss of information and minimize redundancy, we propose two additional principles that apply to collections of classes: completeness, which requires that all properties from the relevant universe be used in a classification, and nonredundancy, which ensures there are no redundant classes. Completeness. Given a relevant universe of instances and properties, every property should be used in the definition of at least one class in the set of classes. If a property is part of the relevant universe, instances possessing it have “meaningful differences” 66 June 1997/Vol. 40, No. 6 COMMUNICATIONS OF THE ACM with respect to those that do not. By requiring all relevant properties to be part of the definition of some class, the principle of completeness supports the “maximum information” aspect of cognitive economy. The principle also supports inference, as it ensures every relevant property can be inferred by considering all classes to which the instance belongs. However, cognitive economy also requires “least cognitive effort,” leading to the next requirement. Nonredundancy. A class that is a subclass of several other classes should be defined by at least one property not in any of its superclasses. Suppose that, in violation of this requirement, a new class is defined by all the properties of some existing classes. Every instance of the new class is an instance of all of the original classes. If there are no additional properties, the new class adds no information about its instances. Therefore, the principle of nonredundancy supports the “minimize the number of classes” aspect of cognitive economy. We define class structure as a set of potential classes satisfying the principles of completeness and nonredundancy; that is: • Every property in the relevant universe appears in the definition of at least one class; and • No class is defined only in terms of the properties of a set of other classes. To have any practical consequence, it must be possible to construct a class structure for any given relevant universe. It can be proven by construction that for every relevant universe a class structure can be formed. Specialization of classes is widely used in both knowledge representation (where it is frequently called subsumption [2, 12]) and in object-oriented design and data modeling [11]. In our propertybased (intensional) view of classes, we defined specialization as forming one class from another class by adding properties. A direct consequence of the observation that distinct potential classes cannot have the same extension is that the instances of a subclass are a strict subset of those of every superclass and therefore are also instances of the latter. The second condition in the definition of class structure also implies that, in a class structure, a specialized class must have properties not possessed by any of its superclasses. To see the intuitive meaning of this, consider again the university example, except that now a research assistant may or may not be a student. If “research assistants who are students” is useful for modeling this domain, we contend there must be properties of this population in addition to those of students and of research assistants, such as a special tuition fee. From a cognitive perspective, without such properties there is no need to define a new class. Properties of Class Structures Class structures have a number of interesting and useful properties for conceptual modeling. First, there will generally be a “pluralism of views” supporting the cognitive principles. This is manifested by two observations: Multiplicity. More than one class structure may exist for a relevant universe having more than one property. Multiplicity implies that several views of a domain, each supporting cognitive economy and inference, can coexist. Inclusion. Every potential class can be included in some class structure. Inclusion ensures that every concept of interest can, in principle, be included in a model that supports cognitive principles. A second useful property of class structures is that they can be made fully descriptive of all instances in the relevant universe. A class structure is required to include all relevant properties. Yet it is possible for a given instance that not all its relevant properties are included in the classes to which it belongs. Also, it is possible for an instance in the relevant universe to be unclassified. In particular, when a class is removed from a class structure, some of its instances may no longer belong to any class. In terms of classification theory, these possibilities reflect

Using objects for systems analysis

nated as a programming and software design discipline offering advantages such as reusability, extendibility and portability. Recently, object thinking has been applied extensively to systems analysis [1, 3, 4, 8, 10]. It has even been claimed that “the real payoff (of the object-oriented approach) comes from addressing front-end conceptual issues, rather than back-end implementation issues” [10]. However, in this evolution, programming concepts have crept into some analysis methods. This is demonstrated by statements such as: “Objects serve two purposes: They promote understanding of the real world and provide a practical basis for computer implementation” [10] or “An object is any thing, real or abstract, about which we store data and those methods that manipulate the data” [8]. We believe that improper adaptation of programming concepts has impeded the successful application of object-oriented systems analysis. Systems analysis involves modeling a domain. It is therefore fundamentally different from software design, which is implementation-oriented. According to Kilov and Ross [7], “the library of generic programming concepts is at too low a level for analysis.” Of course, object-oriented analysis provides input for designing software. Accordingly, software design objectives such as modularity and reuse could be seen as relevant during analysis. We argue, however, that this perspective may interfere with understanding the domain by drawing attention to implementation considerations. This can be detrimental to system success, since “focusing on implementation issues too early . . . often leads to an inferior product” [10]. By drawing on distinctive models of representation we are able to analyze the role of objects in domain modeling. This method leads to a clearer understanding of various aspects of the OO approach. Using Objects for Systems Analysis

What do the pictures mean?: guidelines for experimental evaluation of representation fidelity in diagrammatical conceptual modeling techniques

It is important to articulate the objectives and underlying assumptions behind a growing body of experimental research in conceptual modeling. We provide four guidelines for developing materials for experiments that evaluate conceptual modeling techniques, under the assumption that a primary purpose of conceptual modeling is to facilitate communication between analysts and users in validating domain knowledge during systems development. These guidelines assist in developing experimental materials that support meaningful tests of domain semantics. We present empirical evidence indicating the value of two of the guidelines. We also evaluate selected recent experiments on conceptual modeling with respect to the guidelines.

A framework for targeting banner advertising on the Internet

Constraints that limit accurate targeting of advertising in traditional media may not hold in cyberspace. This paper presents a model for effectively and efficiently targeting hypermedia-based banner advertisements in an online information service. The model takes advantage of information technology to micro-target banner advertisements based on individual characteristics of users. A simple version of the model, which has the virtue of ease of development, is presented. Enhancements are also proposed. These require more effort to develop, but may lead to even more precise targeting of advertisements. Implementation of this framework may benefit both online advertisers and online consumers.

Using cognitive principles to guide classification in information systems modeling

Organizing phenomena into classes is a pervasive human activity. The ability to classify phenomena encountered in daily life in useful ways is essential to human survival and adaptation. Not surprisingly, then, classification-oriented activities are widespread in the information systems field. Classes or entity types play a central role in conceptual modeling for information systems requirements analysis, as well as in the design of databases and object-oriented software. Furthermore, classification is the primary task in applications such as data mining and the development of domain ontologies to support information sharing in semantic web applications. However, despite the pervasiveness of classification, little research has proposed well-grounded guidelines for identifying, evaluating, and choosing classes when modeling a domain or designing information systems artifacts. In this paper, we adopt the cognitive notions of inference and economy to derive a set of principles to guide effective and efficient classification. We present a model for characterizing what may be considered useful classes in a given context based on the inferences that can be drawn from membership in a class. This foundation is then used to suggest practical design rules for evaluating and refining potential classes. We illustrate the use of the rules by showing that applying them to a previously published example yields meaningful changes. We then present an evaluation by a panel of experts who compared the published and revised models. The evaluation shows that following the rules leads to semantically clearer models that are preferred by experts. The paper concludes by outlining possible future research directions.

A Tale of Two Studies: Replicating ‘Advertising Effectiveness and Content Evaluation in Print and on the Web’

ABSTRACT We replicate, using adult web users, a study comparing advertising effectiveness and content evaluation in print and on the web (Gallagher, Foster, and Parsons, 2000). As in the original study involving students, the replication found that advertising was equally effective in the two media. However, while the original study found that evaluation of an article containing advertising was lower when it appeared on the web than when it appeared in print, this result was not replicated. Examination of two subgroups showed that results for the subgroup resembling the student sample were consistent with the original study. We propose conditions under which student samples are appropriate.