Tilburg University, Infolab
P.O.Box 90153, 5000 LE, Tilburg, The Netherlands
firstname.lastname@example.org and email@example.com
There is a lot of literature on graphical and user interface aspects of Web site design [Mau 96, Nie 95, San 96] but little on a systematic way for designing "good" Web sites. Surfing on the WWW shows that a lot of sites are built in a very ad hoc manner and suffer from a number of problems. These problems can be classified according to two perspectives: the viewpoint of the visitor and the viewpoint of the Web manager. The Web manager is usually faced with maintenance problems, the visitors with usability problems (we describe these problems into more detail in Section 2).
A number of researchers have already recognized the lack of design methods for Web sites, or more in general for Web-based information systems, and have proposed methods: HDM [GPS 93] and its successors HDM2 [GPM 93] and OOHDM [SR 95], RMM [ISB 95], W3DT [BN 96], and [TL 97]. The older methods (HDM, OOHDM, RMM) were originally designed for hypertext or hypermedia applications and do not deal with Web-specific issues. Some methods are very implementation oriented (e.g. W3DT). Some have their origin in database design methods and are heavily data driven. These data-driven methods may be able to solve some maintenance problems but they do not meet the usability problems.
We propose a method which is "user centered" rather than "data driven". In a data driven method the available data is the starting point of the modeling approach. In our approach, the starting point is the set of potential visitors (also called audience or users) of the Web site. In our method, these users are classified into user classes and the available data is modeled from the viewpoint of the different user classes. This results in Web sites which are better tailored to their users and will therefore have a higher usability and give greater satisfaction.
Additionally, we make a clear distinction between the conceptual design (which is free from any implementation details) and the design of the actual representation, which takes into consideration the implementation language used, the grouping in pages, the use of menus, static and dynamic links, etc. This distinction is similar to the distinction made in database design between the conceptual schema (e.g. an E-R schema [Che 76]) and the logical schema (e.g. a relational schema [Dat 90]). This distinction has proven its usefulness for more than 15 years. It also allows us to propose a method for Web site design which is not biased by current implementation limitations. And even more importantly, the conceptual schema can be made available to the users. This may prevent them from searching for information which is not available in the site and it may help reducing the so-called lost-in-hyperspace syndrome [Mau 96]. The conceptual schema may also be exploited by search engines to enhance their search capabilities.
In Section 2 of this paper we describe some of the problems Web sites suffer, and indicate to what extent existing design methods deal with these problems. Section 3 gives a general overview of the different phases in WSDM. Section 4, 5, and 6 respectively describe the User Modeling phase, the Conceptual Modeling phase, and the Implementation Design. Section 7 deals with implementation aspects. In Section 8 we presents conclusions and discuss further research.
If the information is not well structured, maintenance problems will occur. These are very similar to the problems which appear in databases: redundancy, inconsistency and incompleteness. This is not surprising as Web sites, like databases, provide large amounts of information which needs to be maintained. The same factors also lead to usability problems, experienced by the audience of the Web site. Information which is repeated over and over again during navigation irritates the user. If some information on the site is found to be inconsistent, the user will probably distrust the whole site. A stale or broken link may be consider as a form of incompleteness, another kind is experienced if the user cannot find the information that, according to his opinion, should be available on the site. Another problem is that of actuality. Organizations and information are often changing so quickly that the information provided on Web sites soon becomes out of date. If a Web site has not been updated for several months, the confidence of the users in the provided information will probably not be very high. An additional problem, in particular experienced by the users, is related to the medium used for Web sites. This medium, Hypertext, requires users to navigate through the information provided. If the navigation process is not well structured or guided, users may easily get lost (the lost-in-hyperspace syndrome [Mau 96]). This makes it more difficult and time consuming to locate the desired information.
An important, but largely neglected aspect of kiosk Web sites is the many different types of visitors they may have. Different types of visitors have different needs. Consider, for example, the Web site of a typical university department. Typical users of such a Web site are candidate students, enrolled students, researchers and university staff.
This shows that different types of users have different information requirements. This should be reflected in the Web site. For example, for a student it should be possible to follow a navigation path which provides him with the desired course and lecturer information and not much more. Usually, this is not the case. For example, at our university site, all information on a lecturer is concentrated on the home page of the lecturer which contains much more, like his research projects, research interest, publications, university functions, etc. This means that the student has to scan the page(s) in order to find the contact hours (if they are given at all).
Different types of users may also share some information requirements. If this is the case, it may be necessary to present the information in different ways to the different kinds of users. As an example we again consider the university site. Candidate as well as enrolled students may need information on courses.
In HDM [GPS 93] the notion of having different presentations for the same content is captured by the concept of perspective. However, perspectives are just a syntactic device to organize information; it has no pre-defined semantics. HDM provides a representation technique to model hypertext applications, but gives little information on how to use it.
OOHDM [SR 95], [SRB 96], a successor of HDM, starts with a "neutral" conceptual schema (using the object oriented modeling technique OMT [RBP 91]). The intended users are taken into consideration in the Navigational Design phase by means of several different navigational models. Although this method considers different types of users, (as far as we know) it does not specify how to describe or distinguish different types of users, and neither does it describe how to derive the different navigational models.
RMM [ISB 95] is based on the E-R method. It allows to split the information concerning an entity into different slices which could be used to support different users. However, the method uses slices only to group information into manageable packages of information.
The method in [TL 97] which focuses on Web-based information systems (WBIS) is also based on the E-R method, but in addition scenario analysis is used to determine how potential users will use the WBIS to accomplish their business goals. Scenario analysis is well accepted for modeling the dynamic aspects of a system (see e.g. [Jac 93], [RPM 95]). For a kiosk site, we think the scenario analysis is less suited. In this kind of applications all dynamic behavior is of the same type: searching information. How a potential user is going to access the Web's information cannot be described in a general way. Trying to do so would result in a long list of all possible ways to reach the information provided. Although a developer's intent may be to guide a user down a series of pages, the actual use may differ. A developer cannot control the links that are made to the Web and so users might enter the Web at any arbitrary point. Most developers will omit scenario analysis when designing kiosk sites.
In our method WSDM (Web Site Design Method, also pronounced as Wisdom) we have opted for a "user-centered" approach. To create effective communication it is not only important to plan "what" to communicate but also to "whom". Therefore, we start by identifying the different types of users. Next, we describe their characteristics and their information requirements. This results in so-called perspectives. Only after this is done, the conceptual design of the Web site starts, taking as input the perspectives and the (possibly already available) business object model of the organization.
Note that a user-centered approach is not the same as a user-driven approach. In a user-driven approach the requirements of the users (acquired by e.g. interviews and scenario analysis) largely guide the design. This is not possible for kiosk Web sites on the Internet because most of the users are unknown and cannot be interviewed in advance. In addition, the main goal of the site is to provide information. The key issue is to do this in such a way that both the provider and the inquirer benefit from it.
In our paper we present a general overview of WSDM. We describe the different phases, how they relate to each other and how they contribute to Web site design. It is not possible to fully describe each phase in detail in this paper. More information can be found in [DL 98].
The method consists of the following phases: User Modeling, Conceptual Design, Implementation Design and the actual Implementation. The User Modeling phase consists of two sub-phases: User Classification and User Class Description. The Conceptual Design phase also consists of two sub-phases: the Object Modeling and the Navigational Design. Software life cycle phases such as Evaluation, Maintenance and Innovation are not considered in this paper. We also do not discuss possible iteration loops. Figure 1 shows an overview diagram of WSDM.
We suppose that the mission statement for the Web site has already been formulated. The mission statement states the purpose of the Web site for the organization. Without giving due consideration to the purpose of the Web site, there is no proper basis for making decisions, or for evaluating the effectiveness of the Web site and the users will have little idea what the Web site is for (i.e. they can only guess and assume).
|Fig. 1. Overview of the WSDM phases.|
If we apply such an environment analysis to our (simplified) department example, we can distinguish the following activities:
The activity "Publicize" is seen as a separate activity because not all publications are the result of research. If we identify the types of people involved in each activity, we arrive at the following user classes:
|Fig. 2. Environment schema for university department.|
In our method, a user class is a subset of the all potential users who are similar in terms of their information requirements. Users from the same user class have the same information requirements. User classes need not be disjoint. The same person may be in different user classes depending on the different roles he plays in the organizational environment. For example, a person can be an enrolled student as well as a staff member.
If within one user class, users have different characteristics which result into different usability requirements, the user class is divide into so-called perspectives, e.g. in our case the user class "Enrolled Students" may be divided into the "Local-Student" perspective and the "Foreign-Student" perspective.
In short, we can say that the user classes describe the information requirements and the perspectives describe the usability requirements. As an example we give the (informal) user class descriptions for the user classes Candidate Students and Enrolled Students.
A User Object Model (UOM) is a model of the business objects in the organization from the viewpoint of the users in a user class. Since a user class usually has only a limited view on the organization, a UOM will only cover part of the total set of business objects of the organization.
In conceptual modeling in general, object models describe the different types of objects, the relationships between these object types (OTs), and rules or constraints on the OTs and relationships. Object-oriented models also describe the behavior of the objects. For our purpose, kiosk Web sites, behavior description is not (yet) needed. Several methods for conceptual modeling exist, e.g. object-oriented methods like OMT [RBP 91], or the more traditional methods like E-R [Che 76] and the Object-Role Model [Hal 95]. These methods are all suitable. Figure 3 shows the UOM for the user class Enrolled Student (in OMT notation).
|Fig. 3. User Object Model for Enrolled Students.|
As explained, a user class may have different perspectives expressing different usability requirements. It is possible that this results in slightly different information requirements. For example, the Foreign-Students perspective requires the same course information as the Local-Students perspective but the information should be given in English. This can be modeled by means of perspective variants. A perspective variant can be compared to a subtype. For each OT in the UOM and for each Perspective, a perspective variant may be given which is a variant of the original OT. For example, for the User OT "Course" two perspective variants may be defined, one for the Local-Students perspective and one for the Foreign-Students perspective. This is illustrated in Fig. 4. For the attributes affected by the perspective, the name of the attribute is followed by the name of the attribute variant, e.g. "Name/English-Name" and "Name/Local-Name". The attribute "Name" in the User OT "Course" is called an abstract attribute, i.e. it has no representation, only its variants do.
|Fig. 4. Perspective variants for Course.|
In this way so-called Perspective Object Models can be built. The Perspective Object Model (POM) for a given perspective is obtained by replacing the User OTs in its User Object Model by the corresponding perspective variants. If a User OT has no perspective variants, the original OT is kept. If a user class does not have multiple perspectives, the UOM has only one POM which is equal to the UOM. In what follows the Object Types of a POM are called Perspective OTs (POTs).
To avoid redundancy when storing the information, and related maintenance problems, the different POMs will be described as views on a single object model. This object model is in fact the Business Object Model (BOM), a conceptual description of the information (business objects) available in the organization. Such a business object model may already exist in the organization. If not, or if it is not specified in a format fit for our purpose, it must be (re-)developed. The classical information analysis methods mentioned earlier could be used for this. POMs are expressed as (possibly complex) views on the BOM.
|Fig. 5. Graphical representation of the navigation concepts.|
Each information component corresponds with a Perspective OT (POT). As a POT may be associated to other POTs (by means of relationships), an information component may contain links to other components to allow navigation to associated information. A navigation component can be seen as a grouping of links. An external component is actually a reference to a component in another site.
To arrive at a Navigational Model, we construct a navigation track for each POM. Such a navigation track consist of three layers. The top layer is the context layer and is used later on to connect the different navigation tracks. The middle layer, the navigation layer provides different ways to access the information. The lowest layer is the information layer and contains the actual information by means of the information components.
We now briefly describe an algorithm to construct a navigational track for a POM. A more detailed description can be found in [DL 98].
Figures 6 and 7 show the navigation tracks for respectively the POMs "Foreign-Students" and "Local-Students". Notice that the navigation layer for Foreign-Students contains less navigation components than the navigation layer for Local-Students. This is because foreign-students do not follow a programme, they only follow courses. Therefore, they do not need to search exams and courses by programme year but only by the course name.
|Fig. 6. Navigation track for Foreign-Students.|
|Fig. 7. Navigation track for Local-Students.|
It should be noted that the design of the implementation may depend on the chosen implementation environment. Limitations of the implementation language (such as HTML) may interfere with this process.
To improve maintainability, more and more Web sites have their underlying information (or parts of it) stored in a database. Pages are generated (usually, but not necessarily, at runtime) by extracting the necessary information from the database. This information may be extracted from existing databases or a new database may be implemented for this purpose. In our method, the basis for such a database design is found in the BOM. The BOM is actually the conceptual schema of the database as it describes the available business objects at a conceptual level. From the conceptual schema, a logical database schema can be generated (using appropriate database development tools), or manually built. The queries needed to extract the information needed for the pages should be easily derivable from the POMs because these are expressed as views on the BOM. More information on linking relational databases to Web sites can be found e.g. in [Gre 97].
Ongoing research aims at refining and elaborating this approach:
In addition to the issues already mentioned it is of course important that the method is extended towards application types of Web sites.
[Che 76] P.P. Chen,
The EntityRelationship Model: towards a unified view of data,
ACM Transactions on Database Systems, 1(1): 471522, 1976.
[Dat 90] C.J. Date,
An Introduction to Database Systems,
Vol. I, 5th edition. Addison-Wesley, Reading, MA, 1990.
[DG 95] J. December and M. Ginsberg,
HTML and CGI unleashed.
Sams.net Publishing, 1995.
[DL 98] O.M.F. De Troyer and C.J. Leune,
WSDM: the Web Site Design Method WiSDoM,
Internal Report, Tilburg University, 1998, in preparation.
[Gre 97] F. Greenspun,
F. Greenspun's Database Backed Web Sites: The Thinking Person's Guide to Web Publishing.
ZD Press, 1997
[GPM 93] F. Garzotto, P. Paolini and L. Mainetti,
Navigation paaterns in hypermedia databases,
in: Proc. of the 26th Hawaii International Conference on System Science, IEEE Computer Society Press, 1993, pp. 370379.
[GPS 93] F. Garzotto, P. Paolini and D. Schwabe,
HDM a model-based approach to hypertext application design,
ACM Transactions on Information Systems, 11(1): 126, 1993.
[Hal 95] T. Halpin,
Conceptual Schema and Relational Database Design,
2nd edition. Prentice Hall, Australia, 1995.
[ISB 95] T. Isakowitz, E. A. Stohr and P. Balasubramanian,
RMM: a methodology for structured hypermedia design,
Communications of the ACM, 38(8): 3443, 1995.
[Jac 93] I. Jacobson et al.,
Object-Oriented Software Engineering,
A Use Case Driven Approach. Addison-Wesley, Reading, MA, 1993.
[Mau 96] H. Maurer,
Hyper-G The Next Generation Web Solution.
Addison-Wesley, Reading, MA, 1996.
[Nie 95] J. Nielsen,
Multimedia and Hypertext the Internet and Beyond.
Academic Press, New York, NY, 1995.
[RPM 95] D. Redmond-Pyle and A. Moore,
Graphical User Interface Design and Evaluation, A Practial Process.
Prentice Hall, Englewood Cliffs, NJ, 1995.
[RBP 91] J. Rumbaugh, M. Blaha, W. Premerlani, F. Eddy and W.
Object Oriented Modeling and Design.
Prentice Hall, Englewood Cliffs, NJ, 1991.
[San 96] D. Sano,
Designing Large-Scale Web Sites.
Wiley Computer Publishing, New York, NY, 1996
[SR 95] D. Schwabe and G. Rossi,
The object-oriented hypermedia design model,
Communications of the ACM, 38(8): 1995.
[TL 97] K. Takahashi, E. Liang, Analysis and Design of Web-based
Sixth International World Wide Web Conference, 1997
|Olga De Troyer received her Ph.D. in 1993 from Tilburg University, The Netherlands. She has held research positions at Control Data's Research Lab in Brussels and at the University of Hasselt (Belgium). Currently, she is an assistant professor at Tilburg University and a visiting professor at the Free University of Brussels. Her research interests include conceptual modeling, object oriented modeling, schema transformations, CASE tools, user interface design and Web site design.|
|Kees Leune graduated from Tilburg University in 1997. Currently he is working as a researcher at the Infolab at Tilburg University. His research interest include object oriented modeling, conceptual modeling and all kind of Internet related activities.|