Supplementing Applications with Hypermedia
Draft Version 1.4 - 8/12/97
DO NOT CITE WITHOUT PERMISSION
Submitted to ACM Transactions on Information Systems

Michael Bieber
Electronic Enterprise Architecture and Design Program; and
Hypermedia Information Systems Research Laboratory,
Computer and Information Science Department
and
The New Jersey Center for Multimedia Research
and
National Center for Transportation and Industrial Productivity

New Jersey Institute of Technology
University Heights-Newark, NJ 07102-1982

bieber@njit.edu
http://megahertz.njit.edu/~bieber
 
 

Abstract

The goal of this research is to provide hypermedia functionality to all information systems that interact with people. Hypermedia focuses both on the relationships among a system's elements and processes, and on how users gain access to them. Hypermedia navigation techniques, by providing direct access to these interrelationships, give users a greater feeling of control within an application, and a greater understanding of its domain, components and results. This research provides system developers with the means to integrate hypermedia functionality into their applications with minimal or no alteration. This paper describes a hypermedia engine which executes concurrently with, but independently of, information system applications, automatically providing their users with both access to the application's interrelationships and supplemental navigation facilities. It presents an open hypermedia systems approach that contributes especially by serving systems which dynamically generate their display contents, which includes most business and engineering analysis applications. Applications remain entirely hypermedia unaware; the engine manages all interrelationship access and navigation on the application's behalf.

Categories and Subject Descriptors
I.7.2 (Automated Hypermedia); H.5.1 (Hypermedia Information Systems)
 

General Terms

design
 
Keywords
hypertext, hypermedia, hypermedia engine, open hypermedia systems, information systems, dynamic mapping, relationship management, navigation, Microcosm, World Wide Web, bridge laws, hypermedia analysis
 

§1 Introduction

In thinking about how we might improve information systems, we can examine applications in several ways. The traditional design methodologies of functional decomposition and data flow analysis help us consider a system in terms of its processes and how information passes through them. An object-oriented viewpoint considers the system in terms of its components and component hierarchies, and the operations applicable to each. Hypermedia considers a system in terms of the relationships among its elements and processes, focusing on how users gain access to them. Hypermedia navigation techniques, by providing direct access to these interrelationships, can give users a greater feeling of control within an application, and a greater understanding of its domain, components and results. We believe that a hypermedia analysis should play a part in the design of every application with user interaction, and hypermedia access should supplement many application feature sets.

This research provides system developers with the means to integrate hypermedia functionality into their applications with minimal or no alteration. This paper describes a hypermedia engine which executes concurrently with, but independently of, information system applications, automatically providing their users with both access to the application's interrelationships and supplemental navigation facilities. Our approach contributes especially by serving systems which dynamically generate their display contents, which includes most business and engineering analysis applications. Applications remain entirely hypermedia unaware [Bieber 1992; Kacmar 1995; Puttress and Guimaraes 1990]; the engine manages all interrelationship access and navigation on the application's behalf. Current hypermedia engines either require an application to incorporate hypermedia functionality directly into its feature set, or provide only manual linking within applications that retrieve and display preexisting information.

We would be remiss not to note the World Wide Web's success in bringing attention to the hypermedia concept. Despite this success, the Web's HTML and XML data models are not as rich as other hypermedia data models and Web browsers generally implement only a rudimentary set of hypermedia features [Bieber et al. 1997]. Most Web applications either require users to create hypermedia links manually within their HTML documents or require the application to be rewritten to incorporate hypermedia links and accommodate the Web interface. Our work presents an alternate approach to hypermedia integration on the Web. Web browsers and the few WYSIWYG Web editors could serve as the user interface system in our engine's architecture described in §3.1.

In terms of open hypermedia systems research (see §2.3), our work contributes a standard generic wrapper for back-end or storage level components-in our case applications which dynamically generate content. (We call these dynamically-mapped information systems or DMISs.) The wrapper automatically maps hypermedia constructs to applications contents. The hypermedia engine delivers hypermedia support based on these mappings. To the extent that other open hypermedia systems support DMISs, they do so on a case-by-case basis instead of in a standard way, they only support limited domains, or they only map hypermedia to the display values as opposed to the objects underlying these display values. §2.3 elaborates on these distinctions.

We begin the next section by examining the interrelationships inherent within application domains. Few applications give users full access to the entire range of interrelationships. We elaborate on the requirements distinguishing dynamically-mapped information systems (DMISs) from display-oriented applications. DMISs dynamically generate content and require automated dynamically-mapped hypermedia support. We then review the current state of hypermedia integration support. Section 3 introduces our implementation architecture and describes our prototype for a mathematical modeling system. Section 4 reviews our future plans and sets a research agenda for hypermedia support of everyday applications.
 

§2 Hypermedia and Dynamic Mapping Information Systems

This section describes how hypermedia can supplement applications and hypermedia's role in dynamic mapping information systems, and reviews other approaches to providing systems with hypermedia support.
 

§2.1 Hypermedia Relationships, Structure and Navigation

The concept of hypermedia grows from a simple but fundamental idea: information is interrelated. The associative link, which so often characterizes hypermedia, embodies this idea by connecting any pieces of information that are somehow related. The sophistication-and potential-of hypermedia lies in applying this idea, i.e., in identifying relationships, items to relate, how users should access them, and how to structure them [F1]. We next discuss these four aspects and show how they can lead to increased comprehension.

(1) Identifying relationships: What characterizes a useful relationship?
Assume the user selects an item of interest in a computer display. Each of the following eight general kinds of relationships represents some aspect of that item. When implemented, it will give the user direct access to it. Of course, different relationships will serve different users such as application developers, managers, analysts and customers. Good implementations will filter and tailor the set of relationships for a given user and task. §3.5 gives examples of each for TEFA, our example DMIS application.
 

Bieber and Vitali [1997] shows how each of these relationships can supplement an on-line sales invoice.

Note that these are general kinds (classes or categories) of relationships one finds in different computational applications. Various hypermedia researchers have come up with different link taxonomies [Trigg and Weiser 1986; Wang and Rada 1995; Oinas-Kukkonen 1996] which contain specific link types for particular applications or domains. These could fall across several of our general link categories. Note too that we have compiled the above set of general link categories based on personal experience in developing and using applications. It is not complete. Neither have we tried to normalize it; a particular semantic link type might fall in more than one category. Nevertheless we have found the current set very useful in thinking about providing relationship management support to analytical computer applications. As part of our future research we plan to come up with a more complete and theoretically robust taxonomy of link categories. Although we have found other general taxonomies [DeRose 1989; Parunak 1991a; Rao and Turoff 1990] inadequate for our domain of generic application support, we might incorporate and use them as a starting point.

(2) Which items should have these relationships?
Taking a "user knows best" viewpoint prescribes making every object on the computer screen selectable-data values, calculation results, labels, titles, report text, entire documents, and so forth-and therefore a candidate for annotation and other relationships. Furthermore, users could access meta-level relationships: every relationship listed above could apply to its own type and to every other. For example, team members should be able to annotate each others' annotations. Users-especially developers within a hypermedia-supported development environment-should be able to annotate a structural relationship or see which processes are interrelated. Similarly, they should have access to all occurrences of any relationship in the system.

Providing such information supports the goal of many organizations to empower lower-level employees with decision-making capabilities. On the other hand, in certain domains developers may deem such access irrelevant or even distracting to the user's task, possibly slowing down time-critical operations. Thus, while a hypermedia analysis helps developers think about an application more broadly, in the end the developer must decide which additional information is (in)appropriate.

(3) How should users access these relationships and the items they associate?The user typically selects an item on the computer screen to follow its link or chooses from a list of available links. (One may have to rank, filter or layer links if their magnitude overwhelms the user.) If a link's purpose or destination is not intuitive, its author or the system builder should label it in some way. Hypermedia access includes navigation, annotation and view-oriented features. Together, these features constitute what we call "full hypermedia functionality," an ideal level of functionality that few of today's hypermedia systems-including our own-achieve. Many systems calling themselves "hypermedia systems," in fact, provide only forward navigation through direct manipulation, and perhaps commenting [Littleford 1991; Rizk and Sauter 1992].

Navigation transports the user among information items. Its features include browsing (following a relationship's link), backtracking, information retrieval-style query based on content, and query based on interrelationships [Halasz 1988; Lee et al. 1996; Bieber and Vitali 1997; Bieber et al. 1997]. Annotation includes comments, bookmarks (hot list items) and user-declared contingent links. In a DSS, for example, annotations can provide justification for a course of action that a decision-maker takes [Bieber 1992]. Users normally traverse from node to node at the detail level, i.e., with each node occupying a window on the screen. View-oriented features enable users to navigate via (graphical) overviews [Landow 1990; Utting and Yankelovich 1989] of the hypermedia network. Overviews help alleviate the network disorientation [Conklin 1987, Thüring et al. 1995] associated with hypermedia's nonrestrictive, user-directed access. View-oriented features also include navigation along recommended paths (trails) [Trigg and Weiser 1986] and guided tours [Marshall and Irish 1989; Garzotto et al. 1996a] of interrelated items. Trails and guided tours both direct and constrain forward navigation. They can document analyses or serve as tutorials, and can be tailored for specific users or tasks. An analyst, for example, could prepare an annotated guided tour to explain a completed analysis or other process. Bieber and Kacmar [1995] describe these hypermedia features in greater detail, especially for computation-oriented applications (i.e., dynamically-mapped systems or DMISs-see §2.2 below), using a geographic information system as the example domain. For a comprehensive overview of hypermedia concepts and functionality see [Bieber et al. 1997; Conklin 1987; Nielsen 1990, 1995].

(4) Structuring Related Information for Comprehension
Hypermedia, as a concept, encourages authors to structure information as an associative network of nodes and interrelating links. This frees authors from the linear, sequential structure that dominates most printed documents. Presenting information as an associative network enables readers to access information in the order most appropriate to their purposes, freeing them from obeying the linear ordering implied within printed documents. Many hypermedia implementations also allow readers to become authors (temporarily, at least) by adding comments (annotations) and additional links among what they read [Conklin 1987]. In all these ways hypermedia promotes options and choice.

In a larger sense, hypermedia increases comprehension [Rana and Bieber 1997; Thüring et al. 1995]. (See Figure 1.) Through the process of structuring information as an associative network (i.e., both creating an organizing structure and then instantiating content into this structure), authors often come to understand that information better. Readers build up a mental model of the information from navigating within the structure and reading content. For readers, comprehension increases through the enriched context that comes from the multiple supplemental relationships which hypermedia encourages authors to provide around a piece of information. Readers build context through browsing these relationships and through other sophisticated navigation support, such as overviews, trails and history support. (Oinas-Kukkonen [1997] extends this framework, considering hypermedia as a technique for enhancing communication between authors and readers.)

Figure 1: Hypermedia as a Concept of Structuring and Access
Hypermedia increases comprehension both for authors who structure information and for readers who navigate within the structure and read its content. Context comes from supplemental relationships and sophisticated navigational features. Link selection can initiate browsing or can begin execution of a process.

Hypermedia structuring extends beyond traditional authored content. Some system developers model processes as associative networks which users invoke through navigating. Representing the process steps as nodes and transitions as links enables developers and users to augment the process with annotations and with links to related information [Noll and Scacchi 1996; Scacchi 1989]. Similarly, some developers structure programs as associated networks controlling which links to activate based on the user's previous interactions (e.g., petri-nets [Stotts and Furuta 1989]). Invoking or performing the process is a reader activity. Authors create programs and processes as part of structuring and instantiating.

While many people only think of hypermedia in terms of enriching authoring and reading environments for manually-crafted documents, hypermedia structuring and access also enhance context and comprehension for environments in which hypermedia functionality is generated automatically. In both cases hypermedia can give direct access to a rich context or web of related information around an object of interest.

---
Two important consequences for systems development follow from the hypermedia concept. First, we can define hypermedia both as the science of relationships [Isakowitz et al. 1995] and as a technology for information access and relationship management. Second, a hypermedia philosophy or vantage point emerges for thinking about applications. As we stated earlier, a hypermedia philosophy helps us think about a system in terms of the relationships among its elements and processes, focusing on how users gain access to them. (Oinas-Kukkonen [1997] discusses many of these issues further.) Thus, hypermedia analysis could supplement the developer's standard design methodology, broadening the scope of objects and relationships a system encompasses. In fact, we believe that a hypermedia analysis should play a part in the design of every application with user interaction. In addition, hypermedia functionality could supplement a system's feature set. This paper does both, applying a hypermedia analysis to our example domain of model management and embellishing information systems, in general, with automated hypermedia functionality.

In the following subsections we describe our focus of dynamically-mapped information systems and describe several techniques for incorporating hypermedia into them. Then in §3 we describe our own architecture and show how hypermedia has enriched the information generated within its dynamically-mapped applications.
 

§2.2 Dynamically-Mapped Information Systems

Many scientific and business applications are computational as opposed to display-oriented [Bieber and Kacmar 1995]. Examples include accounting applications, computer-aided design systems, geographic information systems, expert systems and statistical analysis packages. People use computational applications primarily for their underlying analytic functionality, not for reading or navigating among large amounts of displayable information. Examples of display-oriented applications include encyclopedias, business presentation packages and document retrieval systems. Thus we distinguish between display-oriented applications, which retrieve and display existing information, and those which dynamically generate their display contents at run time [F2]. For the latter we must automatically map hypermedia constructs dynamically, as it usually will be impossible to figure out the location of link anchors in advance. We call any system for which we must dynamically map a hypermedia representation a dynamically-mapped information system (DMIS). Typically display-oriented hypermedia systems support only manual annotative hypermedia linking. Users must select the two endpoints and create a link between them. The hypermedia system then maintains that link so users can navigate it at a later time. But if we automate linking for a display-oriented application (e.g., if its links are too numerous to create manually or if we can infer interrelationships) then we consider these to be DMISs as well. We focus in this paper on DMISs, whether computational or display-oriented.

We model DMISs as having two logical parts: a computational portion and an interface portion. DMIS application content is produced (generated or retrieved) in the computational portion and displayed in the interface portion. As we shall see in §3, the hypermedia mapping takes place as messages pass from one to the other.

DMISs present the special challenge of dynamic regeneration. Suppose an analyst wants to determine the changes in projected profits for different sales levels. He or she performs the analysis and makes comments on each resulting profit calculation. A few days later, when preparing his final report, he needs the calculation results and comments. He wishes to return to them without having to remember the parameter values and then manually re-performing each calculation. Luckily he created a bookmark to each calculation result before discarding them. A bookmark is a one-way link from a globally-accessible menu to something the user wishes to find again-in this case the calculation result. Invoking each bookmark causes the DMIS to reexecute its calculation(s) automatically, and to locate its comment automatically. Bieber and Kacmar [1995] present additional examples of dynamic regeneration for geographic information systems.

To support DMISs we map hypermedia links for a DMIS element based on its internal identity, not to its display value. A stock's price, for example, can change, but the stock's identity never does. Thus we would have to attach the aforementioned bookmarks and comments to the calculation result's internal identifier, not to its display value. This is mandatory for determining what to regenerate, e.g., when the user chooses a bookmark. It also is mandatory for finding link endpoints in displays. There may be multiple instances of the same number in a display with only one of them representing a particular stock's or calculation's value.

During the recent Hypertext'97 Conference, Wendy Hall and Ted Nelson coined the terms flink and flinking, which neatly describe our approach to linking. Flinking stands both for "flinging links" and "filtering links." Flinks are "flung links" or "filtered links." Essentially we have produced a hypermedia engine that takes an application document and throws a set of links on top of it. Thus the links do not reside in documents; they are determined by some external process or agent. Furthermore, lest the set of links flung at the document overwhelm the user, we must filter out the inappropriate ones. Thus one feature that distinguishes ours from the other techniques we discuss in the following section, is that our approach supports domains that generate content in real-time and therefore require dynamic flinking.
 

§2.3 Support for Integrating Hypermedia Functionality into Applications

Fortunately, system builders do not have to program all hypermedia functionality from scratch. There are several approaches to integrating hypermedia functionality into primarily non-hypermedia information systems, several of which may be combined: In related research, Kacmar [1993] compares the toolkit and independent hypermedia engine approaches to coding hypermedia features directly into an application's components from scratch. Oinas-Kukkonen [1997] discusses the difference between integrating hypermedia to support links within a single application versus supporting links primarily to external documents.

Hypermedia data models
Several hypermedia data models, such as the Hypertext Abstract Machine (HAM) [Campbell and Goodman 1988], Dexter [Halasz and Schwartz 1994] and the World Wide Web's HTML provide data structures and guidelines for constructing nodes, links and anchors, and for browsing. Builders can implement one of these instead of making up their own. For example Neptune, a hypermedia CAD application, implements HAM [Delisle and Schwartz 1986]. DHM [Grønbæk et al. 1994] implements Dexter, and RHYTHM [Maioli et al. 1993, 1994] and Wan's hypermedia support for relational database management systems [Wan 1996; Wan and Bieber 1997] implement extensions to Dexter. World Wide Web browsers render HTML. HyTime [International Standards Organization 1992; Newcomb et al. 1991] provides an SGML-based ISO standard model for hypermedia documents (as well as musical documents), although its complexity has encouraged but a few implementations [Buford 1996]. Implementing a data model is no easy task, but it does provide a solid starting point for a hypermedia system.

The toolkit approach for individual applications
The toolkit approach facilitates incorporating hypermedia functionality into individual non-hypermedia applications being built from scratch. An application builder embeds calls to the toolkit's subroutine library in the application's code. Examples of toolkits include Puttress and Guimaraes' Hypertext Object-oriented Toolkit [1990] and the Andrew Toolkit [Sherman et al. 1990]. Anderson [1996] extends the JAVA ATW user-interface toolkit with hypermedia-aware widgets. His approach integrates the toolkit with the Chimera open hypermedia system [Anderson et al. 1994; Anderson 1997] such that the widgets become Chimera clients at run-time. Any interface using these widgets automatically gets a basic set of hypermedia functionality (e.g., make anchor, start link, end link). The application developer then writes event handlers to respond to the hypermedia operations these widgets provide. Garrido and Rossi [1996] are extending the VisualWorks Smalltalk palette to include hypermedia-aware widgets.

Hyperbases
Builders constructing hypermedia systems from scratch need to store and retrieve hypermedia constructs (nodes, links and anchors), which persist between user sessions. Hyperbase storage engines such as HB3 [Leggett and Schnase 1994] provide a data store and management routines for hypermedia constructs akin to database management systems. System builders can integrate hyperbases into their systems, as many link services and open hypermedia systems do.

Link services for linking among applications
Several research efforts have succeeded in creating links among primarily non-hypermedia information systems and in facilitating their traversal. The Sun Link Service [Pearl 1989], HyperBase [Schütt and Streitz 1990], Microcosm [Davis et al. 1992], Multicard [Rizk and Sauter 1992], PROXHY [Kacmar and Leggett 1991], SP3 [Leggett and Schnase 1994] and Chimera [Anderson et al. 1994; Anderson 1997] each has a hypermedia link service that executes concurrently with external systems and provides linking support. Each of the seven hypermedia systems expects client information systems to support anchor creation and selection by embedding hypermedia calls and handling minimal hypermedia functionality in a manner similar to the toolkit approach. The latter five systems also provide multiple levels of hypermedia navigation and annotation support, based on the degree of hypermedia compliance the client non-hypermedia information system provides. Each also provides a limited set of support for client systems that are not hypermedia compatible at all, as we discuss in §3.6. All systems, however, concentrate primarily on supporting manual linking within display-oriented systems.

Generating hypermedia applications from a hypermedia design methodology
Hypermedia design methods present one of the most important developments in the hypermedia field [Bieber and Isakowitz 1995, 1996; Fraïssé et al. 1995]. This involves more than applying well-understood system analysis and design or software engineering techniques to hypermedia. Hypermedia functionality requires new kinds of relationship management and navigation support. Hypermedia design methods provide a systematic approach to relationship management and navigation support, which in turn, should enable consistent, large-scale, robust hypermedia implementations. The most advanced design methodology, RMM [Isakowitz et al. 1995, 1997], has software called RMCASE, which can generate a display-oriented standalone hypermedia application [Diaz and Isakowitz 1995; Diaz et al. 1995]. The designer must model his or her application in terms of entities and relationships; RMCASE then generates a standalone application with the structural and associative links specified in the entity-relationship design (but no others, including no manual linking). Lange [1996] has built software on top of ONTOS for his object-oriented Enhanced Object-Relationship Model design methodology. Other design methodologies such as HDM [Garzotto et al. 1993] and OOHDM [Schwabe et al. 1996] have no accompanying software for generating the application once designed. All these design methods produce new hypermedia applications, complete with their own interface. Our approach differs in that it supports new or existing DMIS applications, without requiring hypermedia-oriented relationships or features to be embedded in the application code.

Hypermedia application software
For standalone applications, many commercial and research hypermedia application environments exist. Here are just a few. The MacWeb design environment produces a hypermedia application for knowledge-based systems [Nanard and Nanard 1995a]. VIKI enables users to build spatial hypermedia applications in which all links are implicit through the user's spatial location of nodes [Marshall et al. 1994; Marshall and Shipman 1995]. These often support only manual hypermedia linking, though some such as KMS [Akscyn et al. 1988] and servers for World Wide Web browsers may be extended through a scripting language.

Independently executing hypermedia engines
Hypermedia engines execute independently of an application with minimal modifications to it, and provide the application's users with hypermedia support. (This differs from but clearly relates to a data model definition of engines-providing "abstract hypermedia data models and storage functionality" to applications [Bapat et al. 1996].) Few approaches provide transparent hypermedia integration as our engine does. Notable projects include Microcosm's Universal Viewer [Davis et al. 1994], Freckles [Kacmar 1993, 1995] and the OO-Navigator [Garrido and Rossi 1996; Rossi et al. 1996]. The Universal Viewer gives users access to a minimal level of Microcosm's hypermedia functionality through the PC Windows operating system. It places selectable buttons in the title bar of the application's windows. Users can select text in these unaware applications and select a Microcosm button to check for "generic" or "keyword" links originating from any portion of the text selected. Links to these applications will launch the application under the Universal Viewer and open the destination document within it.

Kacmar's [1993] work focuses on hypermedia-aware interfaces. His engine works together with the interface to determine how to display objects. The hypermedia engine handles linking and traversal. The interface allows users to select anchors and display the results of traversals. The underlying back-end application must be implemented so that it returns an object's contents when requested. Kacmar's work differs from ours, in part, through its philosophy. His work aims at generating hypermedia-aware interfaces, where our work eventually should lead to hypermedia-unaware interfaces as well as hypermedia-unaware DMISs. Currently the limitations to our approach described in §3.6 prevent an entirely hypermedia-unaware interface, which Microcosm's Universal Viewer and Kacmar's second project [1995] have achieved, at least for non-DMIS applications.

In the second project related to our work, Kacmar [1995] has developed another version of his architecture which operates entirely independently of the interface system. It uses the underlying UNIX twm X Windows system to overlay all hypermedia controls (nodes, anchors, menus, etc.). As with the Universal Viewer, it remains unaware of the application's data model and object identifiers. It uses the window title, as a structured object descriptor, to associate an identifier with the window's contents. Window contents are maintained by the underlying application.

All three approaches seamlessly support an application's other functionality but provide only manual linking-in the Universal Viewer's case on an object's display value; in Kacmar's first system on the object identity; and in Kacmar's second system, on the object's fixed location within a recognized window. Our hypermedia engine, in contrast, contributes by providing seamless, automated supplemental navigation and interrelationship access to dynamically-mapped information systems.

The OO-Navigator comes the closest to our approach, providing a seamless hypermedia support for computational systems that execute within a single Smalltalk environment. They provide a hypermedia layer above the applications which maps application classes to node classes, and relationships among classes to link classes. From these mappings the OO-Navigator automatically generates nodes and links. As mentioned earlier, the OO-Navigator uses Smalltalk's interface modified with hypermedia-aware widgets [Garrido and Rossi 1996]. This approach meets our goal of supplementing Smalltalk applications with hypermedia support without altering them. Our approach applies to both object-oriented and non object-oriented applications.

Open Hypermedia Systems
An open hypermedia system (OHS) combines a link service and a hyperbase, providing linking services to applications [Whitehead 1997; Wiil 1997; Wiil and Demeyer 1996]. Grønbæk and Wiil [1997] present an OHS reference architecture which inherits features from several important OHS models-the Flag Taxonomy [Østerbye and Wiil 1996], the Dexter-based Devise Hypermedia framework [Grønbæk et al. 1994], the Shim Architecture [Davis et al. 1996] and the extended HyperDisco model [Wiil and Leggett 1996]. Figure 2 shows their proposed OHS reference architecture.

Figure 2. A Synthesis OHS Reference Architecture [Grønbæk and Wiil 1997]
This architecture shows the major components of an OHS integration with an application at the user-interface level. The application communicates with the hypermedia linking service through the OHS protocol [Davis et al. 1996]. The hypermedia linking service communicates through appropriate protocols with the hyperbase (hypermedia data base) that stores node, link, anchor and navigation information, and with a document manager that (optionally) stores the content of hypermedia documents and other hypermedia nodes on behalf of the application. The architecture allows for intelligent agent support. It also anticipates future OHS services, as represented by the dashed double bars.

OHSs predominantly provide manual linking services to display-oriented applications. A few claim to support DMISs, but none do this in a systematic manner-all such integration is done ad hoc on a case-by-case basis. For example, in introducing his architectural model for OHS integration, Whitehead [1997] states that "a complete integration consists of a user interface for manipulating anchors and links, and an unbroken path for communication between the {user interface} application and the open hypermedia system." The architectures we propose in §3.1, §3.7 and §3.8, and Chiu's [1997] specialization of them for the World Wide Web provide a systematic way (bridge laws and well-defined DMIS wrapper support) to provide dynamic mapping services for DMIS applications within an OHS. Chiu proposes seven features for OHS support of DMISs: integration with display-oriented applications, integration with DMISs, support for networked distribution, multiple independent users, collaboration, and cross-platform systems, and the ability to extend the OHS (e.g., with a scripting language). He also presents survey results showing how well several OHSs meet these goals.
 
 

§3 An Architecture for Integrating Hypermedia with DMISs

In [Bieber et al. 1995] we introduced a framework for information comprehension and access management (ICAM). It considers information access in terms of information representation, retrieval and delivery mechanisms, describing the ramifications of each on the technical side of an organization, at the managerial level and at the societal level. Hypermedia access is facilitated by appropriate representations. Important aspects of the data representation include granularity of information, and the types of relationships and meta-information available. These determine the types of retrieval or navigational access available. Regarding granularity, because we primarily map hypermedia to identifiable DMIS objects, hypermedia support will only map to the granularity at which objects are specified (see DMIS compliance in §3.6). As we see in §3.5, the hypermedia engine supplements DMISs with information relationship links for whatever metainformation it can access. Delivery considerations such as network capacity and user interface features determine the types of information and navigational features with which the user can interact. Thus the delivery device, while not important conceptually, affects the feasibility and ease of implementing hypermedia functionality. For example, today's major hypermedia delivery vehicle, the World Wide Web, provides only rudimentary support for hypermedia functionality. At the time of this writing, developers must represent all types of interrelationships and construct all advanced hypermedia functionality (e.g., guided tours) from scratch using one simple anchor type and single-step forward browsing [Bieber et al. 1997] (although the new XML standard will allow for multiple explicit semantic link types. As another example, the interface may only permit hypermedia anchors in a window's text, prohibiting users from accessing metainformation and other links to text within window titles or menu labels.

Our research takes the following approach. Hypermedia functionality and interrelationship access supplement the application's normal operations. In terms of information representations, we need to model hypermedia data structures, hypermedia navigation structures, and if not otherwise available, application metainformation and the application internal structure (i.e., its design or schema). All are critical for automated dynamic mapping. In keeping with our goal of altering applications as little as possible-Kacmar and Leggett [1991] call this a "minimal cost" strategy-we do not alter the application's existing data or access structures, but we must be able to interact with them. The mapping, in turn, is critical for retrieving information, i.e., user access. Our engine builds hypermedia functionality on top of the hypermedia data structures using this mapping. The richer the data representation, the more detailed the mapping. This, in turn, affects both the types of navigation available and the types of interrelationships we can infer automatically. The engine uses a relatively simple hypermedia data model consisting of nodes, links and link markers-buttons or anchors. Each carries a rich and flexible set of attributes for capturing the semantic and behavioral aspects of the application elements it represents. Wan has developed a more sophisticated data model for modeling application elements at a higher level and has applied these to relational database management systems [Wan 1996; Wan and Bieber 1997]. We may apply this work in future versions. Bridge laws specify the actual mapping between each type of application element and its equivalent hypermedia construct. Bridge laws are logical rules for mapping two independent representational domains [Kimbrough 1977; Bieber and Kimbrough 1992, 1994], in our case model management and hypermedia.

Mapping and retrieval occur as follows. Our engine intercepts all messages passing from the computational portion of an application to the interface, and uses bridge laws to map each appropriate element of the message to a hypermedia node, link or anchor. Users interact with these hypermedia representations on the screen. If the user selects a normal application command-we map these to the operation relationship links described in §2.1-the hypermedia engine passes the command on to the application for processing. If the user selects a hypermedia engine command (such as a menu command to create an annotation), the hypermedia engine processes it entirely. If the user selects a supplemental schema, process, operation, structural, descriptive, information or occurrence relationship, the engine infers the appropriate application commands, meta-application operations (e.g., at the operating systems level or schema level) or hypermedia engine operations that will produce the desired information. If the user selects a contingent (ad hoc) relationship, the hypermedia engine retrieves the annotation. The interface also handles all other interaction with the DMIS application, such as form-filling and directly manipulating DMIS objects. Thus users access the entire functionality of the DMIS as well as the supplemental hypermedia features through the independent front-end interface system.

Regarding delivery, our engine currently serves applications which can use an independent front-end interface system, such as those which provide an application programming interface (API). We need to use a front-end that can display selectable hypermedia link markers. World Wide Web browsers qualify and indeed we are currently reimplementing our architecture for the Web. As we describe in §4, we plan to extend our engine to existing application interfaces.

This section describes our architecture and our current prototype. We begin in §3.1 with our overall conceptual dynamic-mapping hypermedia engine (DHTE) architecture. §3.2 describes the engine's bridge laws. To jump-start our work, we took advantage of an existing display-oriented hypermedia engine, Microcosm, into which we integrated our DHTE. §3.3 describes Microcosm. §3.4 describes both our integration architecture and our prototype for the model management application, TEFA. §3.5 demonstrates how our prototype provides TEFA with many of the interrelationships described in §2.1. §3.6 describes limitations to our approach. §3.7 gives a glimpse of our final research goal: a multi-user, multi-application distributed DHTE architecture. §3.8 concludes by showing how we might integrate DHTE with other OHS to give them dynamic mapping abilities.
 

§3.1 A Hypermedia Engine Architecture for DMISs

Our research goal is to provide full hypermedia functionality to a dynamically-mapped information system (DMIS) with minimal or no changes to it. Figure 3 presents an architecture to meet this goal. We note that the dynamic-mapping hypermedia engine (DHTE) will serve any DMIS and user interface system (UIS) that has an appropriate wrapper and wrapper knowledge base. To integrate a new DMIS or UIS, one has to specify a wrapper and knowledge base for it. Thus, to provide a DMIS application with hypermedia support, the developer only has to declare the contents of its DMIS wrapper and knowledge base. This may prove straightforward or complex. But in any case it only must be done one time to apply to any DMIS instance. The architecture has the following components.

Figure 3: Architecture for a Dynamic Mapping Hypermedia Engine
Solid lines indicate bidirectional flows of messages among the subsystems. Dotted lines indicate the various instances of (written in) the DMIS, such as a particular expert system within an expert system shell or a program within a CASE system. Dashed lines indicate that a subsystem stores and retrieves information with a database or knowledge base.

User Interface System (UIS)
The user interface system is the "viewer", i.e., the front-end interface through which the user accesses the application and hypermedia functionality.

UIS Wrapper
The UIS wrapper serves three important functions. First it translates messages from the DHTE's standard format to something the UIS can process, and vice versa. Second, it passes messages between the two systems. Third, it buffers the UIS, implementing any functionality the engine requires of the UIS, which the UIS cannot provide itself [Whitehead 1997]. For example, as we discuss in §3.6, the engine requires the UIS to track application elements on the screen. If the UIS does not have this capability, the wrapper may keep track of the location of each element on behalf of the UIS. The term "wrapper" is analogous to "adaptors" and "mediators" in the design pattern literature [Gamma et al. 1995]. Some of the OHS literature calls wrappers "shims".

UIS Wrapper Knowledge Base
The UIS Wrapper knowledge base contains the information the UIS wrapper needs to communicate with its UIS. In it, the UIS wrapper maintains communication formats, current parameter settings and any routines necessary to maintain the level of coordination the DMIS requires of the UIS.

Dynamically-mapped Information System (DMIS)
The DMIS is an application which dynamically generates the contents of its displays, or otherwise requires the hypermedia engine to map hypermedia for it automatically at run time.

DMIS Instances
These are the individual programs, worksheets or whatever other application instances that run under the DMIS system. For example, an expert system shell or relational database management system (RDBMS) would be a DMIS whereas the expert systems written in that shell or the databases manipulated by the RDBMS would be DMIS instances.

DMIS Wrapper
The DMIS Wrapper serves the same three functions as the UIS Wrapper. If necessary, it identifies and marks elements in the DMIS messages to which hypermedia components are mapped [Bieber 1995]. It also defines the routines for determining the interrelationships defined in §2.1. The DMIS wrapper is the key to integrating the DMIS with minimal changes. If the DMIS wrapper can be written to interact with a DMIS's API, then the DMIS will remain entirely unaltered.

DMIS Wrapper Knowledge Base
The DMIS Wrapper Knowledge Base contains the set of bridge laws describing the DMIS' internal structure. It contains network access information for each DMIS so the wrapper can route messages to it and receive messages from it. The knowledge base also contains routines for constructing messages for the DMIS and parsing its responses.

Dynamic Mapping Hypermedia Engine (DHTE)
The engine implements all hypermedia navigation and inference on behalf of the DMIS. The DHTE communicates with all DMISs through a single message format and all UISs with another, single message format. (Currently the open hypermedia systems research community is trying to develop a single protocol for all open hypermedia systems to use [Davis et al. 1996]. As we describe in §3.8, we may adopt this protocol in the future.) The wrappers convert from these standard message formats to ones the DMISs or UISs can use. In [Bieber 1995] we proposed a logical structure for the DHTE consisting of the six knowledge bases summarized as follows. §3.4 lists the actual DHTE modules in this implementation.

Hypermedia Knowledge Base: All hypermedia information registered by users including annotative links, guided tours and trails. The hypermedia engine maintains these independent of the DMIS elements upon which they are based.

§3.2 Bridge Laws

When the engine processes a DMIS message for display, it represents each DMIS component (element or object in that message) with a unique engine component identifier. The engine passes the UIS wrapper the DMIS document with each DMIS component converted to a hypermedia link marker, and for each link marker it passes its display value and engine component identifier. If the user selects a link marker to traverse one of its links, the engine expects the UIS (perhaps helped by the UIS wrapper) to pass the engine component identifier back. The engine uses bridge laws to determine the appropriate links possible for the selected component and returns this link set to the UIS for display. If the user selects one, then the engine guides its traversal, passing the destination content to the UIS wrapper for display. If a contingent link, the engine probably handles the traversal itself, though if the annotation was (to) a non-displayed DMIS instance component, then the engine has to get the DMIS to regenerate it. If any other kind of link, most likely the DMIS or its wrapper will determine the actual destination component-possibly after several calculations resulting in its generation.

Bridge laws describe a DMIS' internal structure adequately to map DMIS instance components to hypermedia constructs. The engine uses three kinds of bridge laws. An identifier bridge law determines what kind of component the user has selected and to which DMIS it belongs. The set of DMIS link bridge laws determines the types of schema, process, operation, structural, descriptive, information or occurrence relationships available for that type of component. Link bridge laws also list any commands which the DMIS or its wrapper should execute to determine the relationship's destination. Each bridge law captures one of these relationships for a type (kind or class) of DMIS component. The engine uses the bridge law specifications to map its relationship as a hypermedia link. The user then chooses among the set of possible links. Upon choosing one, the engine invokes the commands associated with its link bridge law. If a contingent link leads to a DMIS instance component not currently on display, a regeneration bridge law determines the commands and parameters necessary for invoking the DMIS to regenerate it.

The person who knows the DMIS the best-the systems programmer who builds or maintains it-should develop its bridge laws. Because bridge laws are based on the DMIS' internal structure, and each instance shares this structure (e.g., each worksheet written in a particular spreadsheet package shares the same internal structure), the DMIS wrapper will only need one set of bridge laws to serve all of its DMIS' instances. Similarly, all the designs or schemata representing the actual components of each of a DMIS' instances, which the engine uses to determine schema relationships, will share the same internal structure. Thus, once in place, the bridge laws enable the DHTE to map a hypermedia network to any DMIS instance. Instance builders and their users need have no knowledge of bridge laws. To them, hypermedia functionality occurs automatically!

We represent bridge laws and other system knowledge as logical predicates, and we have programmed our engine prototype in Prolog to process them. DMIS wrappers thus must provide the engine with the bridge laws formatted as (perhaps translated into) logical predicates. All arguments beginning with upper-case letters are variables. While Prolog can be used to instantiate (infer the value of) any argument, generally the first one or two arguments already will be instantiated and thus act as the "key" for instantiating the other arguments. While we specify each predicate as a logical fact here, DMIS developers can specify each as logical rule with arbitrarily complex inferencing. As part of our future research, we hope to accept bridge laws specified other than in predicate logic, or at least enable one to construct them in a more builder-friendly bridge law editor.

We now show four Prolog predicates to illustrate our general approach to mapping DMIS application components. §3.4 describes the process in more detail. The hypermedia engine instantiates the following predicate (fact) for each component (document, link marker, link, etc.) which it passes to the UIS for display and stores it in the Active Knowledge Base. The engine's internal engine component identifier is the key to determine which DMIS the component is from and the identifier by which the DMIS knows it.

Based on this internal DMIS identifier, identifier bridge laws determine the type of component selected. Each DMIS component type can have many link bridge laws, including multiple bridge laws for each type of relationship. §3.5 gives the full set of relationships for our sample DMIS, TEFA. Each link bridge law predicate contains the link's label to display, and the command sequence executing the link, if selected. Regeneration bridge laws specify how to regenerate a DMIS component. We have evolved the following set of steps for developing bridge laws. We intend to develop a more comprehensive and detailed set of guidelines as future research. For steps 3 and 4 it may be useful to do a full stakeholder analysis of the DMIS and DMIS instances. One may wish to consider relationships and metainformation that would help an analyst using the system, people who read on-line reports the DMIS instance produces or for which it provides input, the programmers who will maintain the DMIS or DMIS instances, etc. Each stakeholder will find different kinds of relationships useful. Casual users might find the descriptive and process relationships most helpful. Analysts developing a new DMIS instance may find the schema and information relationships most useful in understanding an existing instance and then modeling the new one on it. (The analyst also could benefit from a process relationship on reusing existing instances to construct new ones [F3].) Analysts modifying DMIS instances will take advantage of the occurrence relationships to determine the indirect effects of their changes.

One therefore should plan to filter these interrelationships to the appropriate user. For example, links to parametric information and low level descriptions only should be highlighted for programmers debugging an application or authors considering reusing parts of the application in future applications. Carr et al. [1995] have developed the Web-based Distributed Link Service, which manages different sets of links for different contexts. Users can choose which sets of links to activate. We are considering achieving a first level of filtering by separating bridge laws into separate groups, each representing a particular context. We would activate only certain groups of contexts at any given time according to the user's current role and task. (Brusilovsky [1996] and Calvi and De Bra [1997] use the alternate approach of adaptive linking for activating and deactivating links. We shall consider this approach as well.) Filtering is a high priority for our future research.

Together, our hypermedia data model of nodes, links and link markers, and its bridge laws provide a comprehensive logic-based knowledge representation that enables the DHTE engine to reason about the components (models, data, commands, etc.) of the underlying information systems they map. For further information on bridge laws, see [Kimbrough 1997; Bieber and Kimbrough 1992, 1994]. Also, Wan developed a different model of bridge laws, which focuses on a component's structure and its surrounding components, based upon the Dexter model (see §2.2) [Wan and Bieber 1996; Wan 1996]. Wan's work complements our approach. Its contributions lie in providing a fuller and more formal model of generalized hypermedia, which includes composite constructs.

We currently are revising link bridge laws to be more tailorable and flexible, and to better serve the distributed multi-application environment described in §3.7. Link bridge laws will comprise six sub-specifications. This new format recognizes that the DHTE will be able to do some link determination on its own, and that it might have to consult the DMIS wrapper for more complex link determination. The DMIS wrapper, in turn, may have to invoke the DMIS itself for some link determination.
 

Casting our system in terms of open hypermedia systems (OHS) research, we consider DHTE as a domain-specific OHS where the domain is any DMIS that complies with §3.6's criteria set. The DMIS wrapper and its guidelines provide a standard method for writing a DMIS wrapper. The DMIS and its wrapper correspond to a storage component in OHS designs. However, it constitutes a very special storage component, which can generate its own links and other hypermedia constructs [Chiu and Bieber 1997]. Our contribution lies in developing a systematic way to apply OHS functionality to a major class of business and scientific analysis applications-those that generate their content and displays dynamically.

Now we turn to our prototype. Instead of implementing our architecture from scratch, we built on an existing, non dynamically-mapping hypermedia system. The following subsections explain our approach.
 

§3.3 Microcosm's Architecture

We have implemented our latest DHTE prototype on top of Microcosm [Fountain 1990; Davis et al. 1992]. Microcosm was developed at the Image and Multimedia Research Group at the University of Southampton, England. Microcosm allows users to declare several types of contingent (ad hoc) links manually. Microcosm executes on IBM PCs, Macintoshes and on UNIX, although the PC version is the most advanced and the only one commercialized. Microcosm currently is working on a distributed version of their architecture for the UNIX environment [Hill and Hall 1994]. We decided to integrate DHTE into the PC version of Microcosm, allowing Microcosm both to serve as the UIS and to handle manual linking. The DHTE provided dynamic mapping support for the DMIS, passing documents and dynamically mapped hypermedia constructs to Microcosm for display. We chose Microcosm for some very pragmatic reasons.
  In this subsection we shall describe Microcosm's architecture. In the next subsection we describe our integration and prototype.

Figure 4: Microcosm's Basic Architecture
Messages flow along the arrows. The chain of filters (or functional modules) produces messages for viewers (and other filters further along the chain). Viewers produce messages for the filters. The Microcosm Control System retrieves documents specified in messages from the filters.

Figure 4 sketches Microcosm's architecture. Each viewer displays a particular medium. The Microcosm team has written "fully aware" viewers for media such as text, rtf (converted to formatted text), bitmapped graphics, object-oriented graphics, animation and video. In addition, they have developed macros for Microsoft Word, AutoCad and Visual Basic applications to function as "partially aware" viewers with a reduced set of Microcosm features, and as described in §2.3, to several other "unaware" viewers through their Universal Viewer mechanism [Davis et al. 1994].

The functional modules, called filters, each implement one aspect of Microcosm's functionality. Microcosm's standard filters include:

Microcosm's central control module performs several tasks, including document storage and retrieval, and message passing. All parts of Microcosm communicate exclusively through message passing. Messages pass from the functional modules to the viewers, and from the viewers to the functional modules. Together, Microcosm's independent functional modules and its flexible message passing format make it a very extensible [Halasz 1988] and flexible system. In fact, it is so extensible and flexible that we could add our dynamically mapping engine into this architecture, as we describe in the next subsection.
 

§3.4 Integrating DHTE into Microcosm

Figure 5 sketches our integration architecture. We integrated the entire DHTE system as a Microcosm functional module or filter. Or, from the other point of view, we integrated Microcosm as a UIS for the DHTE. The DHTE Gateway connects our Microcosm wrapper into Microcosm's message passing chain. It converts each Microcosm message from a C language format into a Prolog list, and each message from the DHTE from a Prolog list back into Microcosm's C format. The Microcosm Wrapper maintains all other information necessary to interact with Microcosm. It contains the following routines. Through these, users can interact with the DMIS entirely through the Microcosm UIS.
Figure 5: Integrating DHTE into Microcosm, illustrated with the DMIS "TEFA."
Solid lines indicate bidirectional flows of messages among the subsystems. Dashed lines indicate that a subsystem stores and retrieves information with a database or knowledge base. The TEFA Wrapper Knowledge Base holds bridge laws and other information used by TEFA's wrapper to comply (§3.6) with the engine. TEFA has its own inference engine knowledge base which implements its commands. Dotted lines indicate the various instances of a DMIS, in this case all mathematical models, variables, data scenarios, etc., written within TEFA (§3.4).

The DMIS, TEFA, is a model management system [Bhargava 1990] written in first-order predicate logic. TEFA declarations use the embedded languages approach [Bhargava 1995] in order to ensure maximum metainferencing power from within first-order logic, thereby ensuring robustness and logical completeness. TEFA has the following elements:


model(net_income, 'The simple net income from a pair of revenues and expenses.'). 
modelSource(net_income, 'Michael Bieber').

global_variable(result, 'Net income after taxes').
local_var(result, net_income, net, calculated, '$').
variable_quiddity(result, currency, *, '$').
/* quiddity concerns a variable's units and granularity [Bhargava 1995] */

global_variable(revenues, 'Gross receipts, i.e., money taken in').
local_var(revenues, net_income, rev, given, '$').
variable_quiddity(revenues, currency, *, '$').

global_variable(expenses, 'Expenditures').
local_var(expenses, net_income, exp, given, '$').
variable_quiddity(expenses, currency, *, '$').

global_variable(gross_profit, 'Revenues less expenses-net income before tax').
local_var(gross_profit, net_income, gross, calculated, '$').
variable_quiddity(gross_profit, currency, *, '$').

global_variable(tax_rate, 'Flat tax rate').
local_var(tax_rate, net_income, tax, given, none).
variable_quiddity(tax_rate, null, *, none).

equation(net_income, 1, net =: gross * (1 - tax)).
equation(net_income, 2, gross =: rev - exp).

assumes(net_income, 1, 'There is a single flat tax rate used.').
assumes(net_income, 2, 'There is a single pair of revenues and expenses').

Table 1: The Net Income Model written in TEFA
 

Table 1 presents a sample "net income" mathematical model in TEFA. Each TEFA element has several operations available. For example, for a mathematical model, the user can describe it, determine which data scenarios have data for executing it, execute it, list its submodels or list its local variables. For a local variable, the user can describe it and determine which models use it. TEFA's inference engines perform these operations and reasoning. Figures 6 and 7 show screen shots of this model. §3.5 presents the TEFA bridge laws declaring relationships inferable for it (or any other mathematical model instance in TEFA).

Figure 6: Screen Shot from the Prototype Showing All Available Links for TEFA's "net_income" Model.
In response to selecting the "net_income" anchor, the engine displayed eleven navigation choices for this mathematical model. The net_income model has four available commands to create comments and bookmarks, and seven possible links. Each link was generated from a separate bridge law. Within the available links dialog box, the user now is selecting the "Run Model" operation link. All anchors in documents are underlined.

Figure 7: Screen Shot of TEFA's Execution of the "net_income" Model
Choosing the "Run Model" operation link in Figure 6 caused TEFA to execute the "net_income" model. TEFA returned the report "run(net_income,flows(1))," to which the DHTE engine dynamically mapped the underlined anchors.

TEFA, as a DMIS, communicates with the DHTE entirely through messages generated by or received by the TEFA wrapper we wrote. TEFA stores all declared elements in the element instance base.

The DHTE converts TEFA messages into text documents, with the position of each link marker noted. TEFA messages consist of plain text, structured text (lists, tables, etc.) and marked-up TEFA components. The mark-up identifies which components have potential relationships and gives the display value and internal TEFA ID for each. TEFA's bridge laws, held in its wrapper's knowledge base, specify the component types, and the operations and other relationships available for each. We elaborate on these in the following subsection. TEFA's wrapper has the following routines:

In the rest of this subsection we present some lower-level technical details on the prototype's integration and interaction.

The DHTE is event driven, responding to user activity. Each DHTE module described below responds to and/or produces a Microcosm message. Microcosm messages consist of an arbitrary number of attribute/value pairs representing each message argument. The "action" attribute/value pair represents the message type. The DHTE passes an action parameter in its messages declaring menu items, anchors and links. When the user selects one of these, Microcosm returns a message with that action type. The DHTE includes the following modules. For each we show the action attribute's value and relevant action parameters stored with menu items, anchors and links for the Microcosm message that invokes the module's functionality, and each message that DHTE returns in response. We write standard Microcosm parameters in uppercase and specialized DHTE parameters in lowercase [F4].
 

  • Module: Initialize DHTE
    •  
    message from  action value  actual action parameters 
    Microcosm (DCS)  SYSTEM.INFO  {none}
    DHTE {no direct response} 
     
     
  • Module: Initialize Menus
    •  
    message from  action value  actual action parameters 
    Microcosm {no direct call} 
    DHTE ADD.MENU.ITEM  dhte_menu_selected  
    (Owning_System, Command) 
    where Owning_System is the system that will execute the action (DHTE or DMIS) and Command is the actual command that system will execute
     
     
  • Module: Menu Selected
    •  
    message from  action value  actual action parameters 
    Microcosm  
    (Viewer)     		 dhte_menu_selected 
    (Owning_System,  
    Command)     		 {none}
    DHTE {no direct response} 
     
     
  • Module: Create Document
    •  
    message from  action value  actual action parameters 
    1. Microcosm  
    (Viewer)     		 a menu or link selection  
    listed elsewhere     		 {none}
    2. DHTE IMPORT.DOCUMENT  {none} 
       
    message from  action value  actual action parameters 
    3. Microcosm (DCS)  DOCUMENT.IMPORTED  {none}
    4. DHTE AUTODISPATCH {none} 
       
    message from  action value  actual action parameters 
    5. Microcosm  
    (Viewer)     		 FIND.BUTTONS  {none}
    6. DHTE FOUND.BUTTON dhte_button_id  
    (Component_Identifier);  

    dhte_button_owner 
    (Owning_System) 
       
  • Module: Regenerate Document
    •  
    message from  action value  actual action parameters 
    Microcosm (DCS)  DISPATCH or AUTODISPATCH  {none}
    DHTE AUTODISPATCH  {none} 
     
     
  • Module: Anchor Selected
    •  
    message from  action value  actual action parameters 
    Microcosm  
    (Viewer)     		 FOLLOW.LINK  dhte_button_id  
    (Component_Identifier);  
    dhte_button_owner 
    (Owning_System) 
    DHTE DISPATCH  dhte_link_command  
    (Command_Sequence)  
    dhte_link_command_owner 
    (Owning_System) 
     
     
  • Module: Link Selected
    •  
    message from  action value  actual action parameters 
    Microcosm  
    (Bounce)     		 dhte_link_action  dhte_link_command  
    (Command_Sequence)  
    dhte_link_command_owner 
    (Owning_System) 
    DHTE {no direct response} 
    Other DHTE routines create annotations, display annotations, get user input from a user dialog box, and close DHTE.
     

    We now describe DHTE and Microcosm's interaction with our example DMIS, TEFA.

    This triggers a process similar to that just described for this new command. The document in Figure 7 shows TEFA's execution result with all objects highlighted as anchors by the DHTE. [Kimbrough et al. 1990; Bieber 1992, 1995; Bieber and Kimbrough 1992] present additional examples which could be implemented in this architecture.
     

    §3.5 Hypermedia Relationships for TEFA

    Table 2 lists TEFA's bridge laws giving users direct access to various aspects of the TEFA application. These aspects reflect six of the eight general kinds of relationships for applications discussed in §2.1. (We discuss the others below.) Most of the bridge laws just give users direct access to the standard TEFA commands by clicking on an anchor (highlighted instance of a TEFA element) instead of typing the command in TEFA's standard command line interface. Each bridge law includes the TEFA command that users previously had to type manually to invoke it. In addition, we added six supplemental relationships as bridge laws, each including the sequences of TEFA commands necessary to generate its result. TEFA's bridge laws reside within the TEFA Wrapper Knowledge Base shown in Figure 5. They function for all instances of the nine TEFA elements listed in the table.

    Element Type
    Relationship
    Type
    Bridge Law Description
    Supplemental
    Bridge Law?
    calculation result operation explain result
    calculation result operation convert units (quiddity)
    datum information list source
    datum operation convert units (quiddity)
    datum operation change value
    equation description describe equation
    equation information list equation assumptions
    equation occurrence show this equation in other models
    yes
    function information describe function
    function operation execute this function
    global variable information describe variable
    global variable occurrence list data scenarios containing this variable
    yes
    local variable description describe variable
    local variable structural list models using this variable
    model description describe model
    model operation execute (run) model
    model schema suggest data scenarios for model
    model structural list local variables within model
    model schema list models in the same application domain
    yes
    model structural list variables in the model of a given quiddity
    yes
    model information generate statistics about this model
    yes
    relation description describe relation
    scenario description describe data scenario
    scenario schema list models that can use this data scenario
    scenario information generate statistics about this data scenario
    yes
    Table 2: The TEFA Wrapper's Bridge Laws

    Following are some notes on the general relationships and bridge laws.

    Operation Relationships
    Because all the non-supplemental bridge laws in Table 2 actually invoke existing TEFA commands, we could have labeled each as an operation relationship. We have chosen, though, to categorize many of them in terms of the other general relationships types. Doing so pointed out to us several "under supported" general relationship types. This prompted us to follow our own advice from §2.1. Using the general relationship types to prime our thoughts, we (surprisingly quickly) thought of many individual relationships that TEFA lacked, from which we chose six to implement preliminarily. We have checked these six in Table 2.

    Schema Relationships
    To demonstrate supplemental schema relationships, we added an external set of application domains to TEFA's wrapper and allow model builders to associate their models with an application domain.

    Process Relationships
    TEFA does not provide process support and we have not yet supplemented TEFA with any. We do, however, envision bridge laws that declare step-by-step processes such as adding new models or performing a complete analysis that considers an adequate number of different data scenarios (i.e., adequate what-if analysis). We leave this to future research.

    Contingent Relationships
    Through bridge laws, DHTE can provide users with the other seven general relationships automatically. In addition, hypermedia as a concept inherently includes uninferable contingent or ad hoc relationships declared by users. For example, users could create:

    Note that DHTE supports two different orders of annotation, as reflected by two types of comments and two types of bookmarks in Table 2's DHTE bridge laws. One is to an object occurrence in a specific node or document. The second is to an object in general, i.e., not associated with any particular occurrence. But how should we display a general object without any associated manifestation of it? As originally reported in [Bieber 1990], we take the same approach as Grønbæk and Trigg [1992], namely representing an object by its relationships. Selecting, for example, a general bookmark to TEFA's "profit" model leads to the set of all available links for that object in general, displayed in Microcosm's available links dialog. Note that annotations on DMIS objects are stored in DHTE's Hypermedia Knowledge Base (§3.1). DMISs such as TEFA have no access to or cognizance of these annotations.

    ---
    The above items illustrate our general relationship types for a mathematical modeling DMIS. Wan [1996] describes some associative, operation, structural and metainformation relationships for relational database management systems. Bieber and Kacmar [1995] describe some useful relationships for geographic information systems. Bieber and Vitali [1997] describe general relationships for a customer invoice system. Although they do not classify them as such, Kimbrough and Vachula [1996] provide a set of analogous schema, structural, information and occurrence relationships for object oriented programming environments. This set would make an excellent collection of supplemental bridge laws for that domain.

    We note the limitation that DHTE currently does not support (meta)relationships on other relationships, such as annotating an occurrence relationship, or providing metainformation on process relationships. The following subsection describes the compliance DHTE demands of DMISs and UISs to support some of DHTE's other limitations.
     

    §3.6 Compliance

    Our hypermedia engine requires a specific degree of cooperation from both the user interface systems and from DMIS applications. In contrast, Microcosm supports three levels of user interface systems [Davis et al. 1994]. Fully-aware viewers have been written or altered to comply in full with Microcosm's messages, giving users access to all Microcosm menus and features. Partially-aware viewers such as Microsoft Word comply in part with Microcosm through macros. Unaware (or "uncooperative" [Kacmar and Leggett 1991]) viewers provide no Microcosm features. Microcosm supports these through the Universal Viewer described in §2.3. (Davis et al. [1994] actually list six levels of integration, each for a different level of compliance: tailor-made viewers, source code adaptation, object-oriented reuse, application interface level adaptation, shim or proxy programs and launch only viewers.) DHTE's user interface limitations arise primarily because we map hypermedia links to DMIS elements based on their internal identity, not to their display values. DMIS compliance requirements arise from the need to intercept DMIS messages for display and to identify DMIS elements within these messages so the DHTE can infer all hypermedia relationships for each. Kacmar and Leggett [1991, Puttress and Green [1990] and Rizk and Sauter [1992] also investigate compliance requirements. Isakowitz [1993] and Isakowitz and Stohr [1992] report on an integration architecture using state-change messages that claims to require less coordination among the hypermedia engine and its external systems.

    In this subsection we describe the compliance the DHTE demands from the UIS and the DMIS. This compliance, together with the functionality described in §4 which we have not yet implemented, constitute the limitations to our approach and to our current prototype.
     

    User Interface System Compliance

    Either the user interface system or the user interface wrapper must provide the following support to integrate with the DHTE.
     
  • Provide hypermedia prompts

    • The interface must distinguish hypermedia anchors in some way from other window content. The interface must distinguish anchor selection from other user actions, thus providing a mechanism to invoke link traversal. If users can edit the hypermedia anchors, then the system must distinguish selection for editing from selection for traversal.
     
  • Menus

    • The interface must display DHTE and DMIS menus in some fashion, though this could be in some crude manner such as making each menu a hypermedia anchor, which would be placed in all documents.
     
  • Identifying objects in UIS workspaces

    • If users may edit UIS documents, then the interface must track hypermedia anchors when their positions shift, so that users still can select and traverse them. (A few researchers are tackling ways to find anchors once text has shifted [Davis 1995a, 1995b].) Kacmar [1993] also notes the need for coupling a call to windowing systems with a call to an underlying interface component to obtain the object identifier.
     
  • Passing information to the DHTE

    • When the user selects a menu or selects an anchor for traversal, the interface must inform the DHTE.
     
  • User dialogs

    • The interface must support some kind of forms or dialog box to collect user input. TEFA sometimes requires users to input parameter values, such as the name of the data scenario, over which to execute a model.
     
  • Accepting displays from the DHTE

    • In most information systems users create documents manually. With a hypermedia engine, the UIS must accept the externally-generated documents with embedded objects that the DHTE passes. Ideally the UIS will handle dynamic changes as well. The engine may add additional objects to open documents (e.g., when users create their own annotations on the front-end workspace [Pearl 1989]). Dynamic updating, such as with hot linking [Meyrowitz 1989; Catlin et al. 1989, Newcomb et al. 1991; Bieber et al. 1997] may change the display value of anchors [Isakowitz and Stohr 1992], such as displaying the current price of a stock. This requires the UIS to accept external interrupts so it can accept information not caused by (or anticipated in response to) a user interface event.
     

    DMIS Compliance

    Either the DMIS or the DMIS wrapper must provide the following support to integrate with the DHTE. This list closely resembles compliance sets identified by other OHS researchers [Davis et al. 1994].
     
  • Communicate with the DMIS

    • The DHTE must intercept communications between the DMIS and its interface so that DMIS messages are routed through the DMIS wrapper to the DHTE. The DHTE then can map hypermedia link anchors to it. The more loosely coupled (the more independent the computational and interface portions) and modular an information system is, the simpler hypermedia integration will be.
     
  • DMIS message markup

    • The DHTE must identify DMIS objects in messages in order to determine whether to map hypermedia anchors to each. Currently the DHTE relies on the DMIS wrapper to identify and mark up each DMIS object in some way, so the DHTE knows where it lies in the message and what kind of object it is. This limitation could be alleviated for object-oriented DMIS if the DHTE can query the DMIS objects embedded in messages. Otherwise the DHTE would have to rely on some kind of sophisticated content analysis. An advanced CS could employ, for example, sophisticated lexical analysis techniques to infer undeclared keywords. The DMIS also could provide some content analysis routines for interpreting its messages.
     
  • Unique IDs:

    • Because the DHTE maps contingent annotations and other links to DMIS object identifiers, the DMIS needs unique identifiers on each object the DHTE could turn into a link anchor. This includes DMIS documents. In TEFA, for example, the TEFA wrapper had to assign and maintain unique identifiers for each document, as TEFA itself did not.
     
  • Builders must write bridge laws

    • The person who knows the DMIS best-the systems programmer who builds or maintains it-should develop its bridge laws. The information system builder must be both willing to and capable of developing a set of bridge laws that accurately captures the structure of his system. The DMIS must have a robust, inferable structure in order for the builder to capture it in bridge laws. Once in place the bridge laws should map a hypermedia network to any of the system's application instances-documents, worksheets, program, etc. (Instance builders and users need have no knowledge of bridge laws. To them, hypermedia functionality occurs automatically!)

    Currently builders must represent bridge laws in Prolog predicates. We hope to remove this restriction by accepting other formats, perhaps through a bridge law editor. (Wan [1996] and Wan and Bieber [1997] have declared templates for bridge laws, but these too require complex specifications.)

    Each builder must develop his own set of bridge laws. We hope to develop bridge law libraries that map classes of information systems-complete "standard" bridge law sets that handle the models, attributes, data and operations found, e.g., in linear program (LP) packages, spreadsheet packages, or rule-based expert system shells. The builder of, say, a new LP package would only have to match the elements in his system to those in the standard LP set. The standard set would provide most of the bridge laws for his system. This would reduce the builder's effort both in determining which kinds of bridge laws would represent his system adequately and in developing these laws. Wan [1996] and Wan and Bieber [1997] have demonstrated this approach for relational database management systems.

    ---
    Even if a DMIS has no bridge laws and passes messages without objects, the hypermedia engine still will provide standard hypermedia functionality (user annotation, backtracking, etc.) In this case the user will not be able to access DMIS items or operations in a hypermedia fashion through the DHTE.
     

    §3.7 The Future: A Multi-user, Multi-Application Distributed Architecture

    To fully support the broad range of information systems found in organizations, the hypermedia engine must support multiple applications and users. We envision that hypermedia even could integrate independent applications for interrelated tasks and processes [Bieber 1996]. Consider for example, the query "how many trucks travel on which routes between Newark, New Jersey and Philadelphia, Pennsylvania?" The freight databases currently available only contain commodity flows among regions in tons per item. To answer, the analyst normally would (1) extract truck tonnage information from a series of these databases, (2) use an analysis program to convert tonnage to numbers of trucks based on a commodity density analysis, (3) use a second analysis program to determine which routes trucks typically travel, and (4) generate the final report for display on a word processor. A hypermedia engine could manage the task for the analyst (as a process relationship), passing appropriate outputs of prior applications as inputs to the current one when the user invokes it. At any point, for any component, the user could see the eight general relationships discussed in §2.1 and §3.5, including for example, how a particular datum is processed in each of the four independent applications.

    The architecture we propose in this section could provide this environment. Figure 8 describes a distributed, multi-user, multi-application architecture at a high, conceptual level. All components may be distributed, including the DHTE and the Master Schema Knowledge Base (MSKB). The DHTE and MSKB can be replicated (have separate implementations) for independent groups of users.

    Figure 8: High Level Architecture of a Distributed, Multi-user, Multi-application DHTE.
    All components, including the DHTE and Master Schema Knowledge Base may be distributed.

    The MSKB plays a crucial role in this architecture. This knowledge base ties together the various DMISs in the system. We intend for it to contain meta-bridge laws which coordinate the bridge laws among DMISs. This will enable process support across applications [Tanik et al. 1996], occurrence relationships (locating and tracking a piece of data across applications) and schema relationships (e.g., finding the corresponding models or routines) across DMISs. Our tentative approach is to develop an application design coordination schema which will allow us to map standard design documents from different design methodologies (entity-relationship diagrams, data flow diagrams, OMT design diagrams, etc.) to each other. Kimbrough and Vachula [1996] provide a preliminary logic-based approach for object-oriented design methodologies, which we intend to expand for other methodologies.

    Other hypermedia research systems either support multiple users or multiple applications. Yet to our knowledge, no other hypermedia engine besides OO-Navigator for Smalltalk applications [Garrido and Rossi 1996] provides full hypermedia support to computation-oriented (dynamic mapping) applications with minimal or no alterations. The open hypermedia systems field has made progress towards seamless integration, but has produced no standard way to support business and engineering analysis applications requiring dynamic mapping.
     

    §3.8 Integration with other Open Hypermedia Systems

    We also could extend out architecture to integrate the DHTE with other open hypermedia systems (OHSs). Figure 9 shows this potential integration. We most likely would connect the DHTE to the OHS through the Open Hypermedia Protocol [Davis et al. 1996]. (Within Grønbæk and Wiil's [1997] OHS reference architecture from Figure 2, the DHTE nominally would reside at the storage layer as it passes complete documents and other node content to the OHS for display.) Much as in our integration with Microcosm, we would rely on the other OHS to handle all user interface interaction. In return, the DHTE would supplement the OHS with dynamic mapping services. This would enable the OHS to give users access to analytical applications and other DMISs.

    We also could take advantage of this arrangement to achieve a division of labor. The OHS would provide all standard hypermedia navigation and linking features. We would provide all dynamic mapping and DMIS access. The integration would enable us to concentrate our future research efforts on dynamic mapping abilities and relationship management. The OHS community will continue to research integration issues, and continue to improve hypermedia structures, annotation and navigation features. Each of us would be able to take full advantage of the research successes of the other on an on-going basis.

    Figure 9: Architecture for Integrating DHTE with Other Open Hypermedia Systems
    Starting with Grønbæk and Wiil's [1997] Synthesis OHS Reference Architecture in Figure 2, we connect the DHTE hypermedia engine to the Hypermedia Service through the Open Hypermedia Protocol at the storage layer.

     

    §4 Hypermedia and Information Systems: A Research Agenda

    Several visionary hypermedia articles have called for "industrial strength" hypermedia support for the everyday applications used in organizations [Engelbart 1990; Malcolm et al. 1991; Parunak 1991b]. Various authors and workshops have defined a core set of hypermedia features to which users should have access within those systems [Ashman et al. 1994, 1996, 1997; Bieber and Kacmar 1995; Bieber et al. 1997]. Research into open hypermedia systems has made a start towards this vision, but primarily for manual hypermedia support or for applications requiring little or no dynamic mapping support. Our work takes the next step, providing supplemental hypermedia support to hypermedia-unaware applications with minimal changes in a general manner.

    We plan to make several improvements to our prototype. First, we intend to implement the full range of hypermedia functionality including sophisticated backtracking [Bieber and Wan 1994; Garzotto et al. 1996a], metarelationships on relationships, guided tours and overviews. Second, we plan to support process relationships [Bieber 1996; Bieber and Vitali 1997; Scacchi 1989; Noll and Scacchi 1996] much more fully, both for single applications and to support analysts working with multiple applications (as in §3.7). After this we intend to extend our hypermedia engine so it works with a DMIS' existing interface. In this case the UIS in Figures 3 and 8 would be the current interface of its DMIS. The engine would intercept the application's normal communications and insert hypermedia components into these messages [Bieber 1995]. In addition we plan to develop full guidelines for DMIS integration to expand those in §3.2. We also plan to implement the multi-user, multi-application architecture of §3.7 for DMIS applications, and to integrate our DHTE with existing OHS as described in §3.8. We also hope to investigate providing hypermedia support through hypermedia-unaware UISs, thereby complementing our approach to supporting DMISs. Finally, we note our firm commitment to demonstrating supplemental hypermedia support to other DMIS applications beyond model management and relational database management systems, etc. We would welcome collaboration with others in applying hypermedia to additional application areas such as CAD, computer-based training and education, CASE, EIS, expert systems, GIS, spreadsheets, and others.

    At a larger scale we are witnessing the development of the new field of organizational hypermedia. Organizational hypermedia concerns the effects and implications of hypermedia as a concept and as a technology for supporting individuals and modern organizations. Organizational hypermedia [Isakowitz 1993] focuses on an organization's data and information management, as well as on managing collective knowledge [Marshall et al. 1995; Nanard and Nanard 1995b], organizational memory [Ackerman 1994; Croasdell et al. 1997; Tervonen et al. 1997] and decision support [Carlsson 1997; Chang et al. 1993; Mao et al. 1996]. Bieber and Isakowitz [1996] ask four fundamental questions for organizational hypermedia:

    - How can we give users better access to information?
    - How can access to and management of relationships enhance existing applications?
    - How can one develop hypermedia-supported applications?
    - How might one evaluate the effectiveness of hypermedia-assisted solutions?

    While our work addresses aspects of the first three questions to a greater or lesser extent, to truly meet the challenge posed by the industrial strength vision of hypermedia support, much remains for investigation. First, we need more evidence of how hypermedia can support everyday users and their applications, in order to convince software developers to incorporate supplementary hypermedia support. Second, we need robust software engineering and systems analysis and design methodologies for developing hypermedia support. The methodologies mentioned in §2.3 constitute an excellent start. They support only certain types of applications and development from scratch. We need a design methodology for DMIS-style applications. We need design approaches for retrofitting supplementary hypermedia support to existing applications (including an organization's legacy systems [Bennett 1995]). We also should reiterate our belief from §2.1 that a hypermedia analysis can supplement other design methodologies. As part of our future research, we intend to develop a full hypermedia philosophy analysis methodology for non-hypermedia applications.

    Regarding organizational hypermedia's fourth question, we note the paucity of metrics and evaluation approaches for hypermedia-supported applications [Brown 1990]. No self-respecting programmer or developer would release a commercial-quality product without thoroughly testing it first. Furthermore, in an effort to convince developers of hypermedia's usefulness, judging its effectiveness requires effective evaluation of hypermedia-supported systems. One can view evaluation in two ways. One can evaluate the actual node and link structure within an application (e.g., Do all links have endpoints? Can users get to the information they want within n link traversals?) One also can evaluate general accessibility or how well an application follows the hypermedia spirit or the "hypermedia feeling" [Vanharanta et al. 1995]: For every object visible on the screen, can the user easily gain access to all related information? Can the user clearly "see" how the information is related and under what context? Do the relationship and context remain clear once the user views the related information? Do the users of a hypermedia-enhanced system understand the system, its components and its results, and have confidence in these? A few researchers have started to work on hypermedia metrics [Botafogo et al. 1992; Garzotto et al. 1995; Rivlin et al. 1994; Vanharanta et al. 1995]. Commercial companies that provide hypermedia conversion and organizational support have developed robust sets of automatic evaluation tools (e.g., [Johnson 1995]) but these remain largely unpublished. Some researchers have done evaluation studies [Brown 1990; Minch and Green 1996; Vanharanta et al. 1995] but few actually advocate an evaluation methodology [Garzotto et al. 1995].

    The hypermedia field contains woefully little research and tools supporting analytical information systems. Yet this is where we believe hypermedia will have its greatest impact, especially for the scientific and business worlds. With the increased consciousness of hypermedia which the World Wide Web has brought, developers and the organizations for whom they work have never been more open to considering their computer applications in a new light. Yet few guidelines exist for assisting developers in providing access, i.e., in deciding what to link, and few tools enable easy hypermedia integration for the vast range of applications that dynamically generate information. And thus, while hypermedia as a concept has made major contributions to the fields of collaboration, technical communications and literature, and while both outstanding research-oriented and commercially-available hypermedia environments exist, the mainstream information systems development and software engineering fields, for the most part, have not taken advantage of hypermedia concepts. The time is ripe for an explosion in hypermedia support. We invite researchers in all related fields to help guide the current intense interest to fruition.
     

    Acknowledgments

    Many people have contributed to this research in many ways. Foremost I am indebted to the Microcosm team at the University of Southampton for hosting a fortnight-long visit without sleep, during which much of the prototype was crafted. Wendy Hall, Hugh Davis, John Russel, Stu Goose and Simon Knight all provided remarkable assistance. Ian Heath was my main Microcosm guru. He put in loads of overtime coding the Microcosm and Visual Basic parts of the Microcosm UIS wrapper. I am most grateful for his help. Chao-Min Chiu of National Institute of Technology, Kaohsiung, Taiwan also put in long hours in development support. Hemant Bhargava of the Naval Postgraduate School developed TEFA and helped in the brainstorming that led to the original forerunner to this paper's prototype. Ken Anderson of the University of California at Irvine, Uffe Kock Wiil of Aarhus University in Denmark and V. Balasubramanian of E-Papyrus, Inc. made invaluable and insightful comments clarifying the focus of this research. Alejandra Garrido of the University of Illinois at Urbana/Champaign, Scott Moore of the University of Michigan and Steve Roehrig of Carnegie-Mellon University also made extremely useful comments on prior drafts. Tomás Isakowitz and Steve Kimbrough of the University of Pennsylvania have participated in numerous discussions underlying this research's concepts. Sharman Lichtenstein of Monash University, Paul Thistlewaite of the Australian National University and Murray Turoff of the New Jersey Institute of Technology (NJIT) all have contributed to expanding the set of general hypermedia relationships. I very much appreciate all of their useful comments. Praveen Ramanathan at NJIT has been very helpful in the beginning stages of designing the follow-on architecture mentioned in §3.7, which has lead to several improvements in the text of this article.

    I gratefully acknowledge funding for this research by the NASA JOVE faculty fellowship program, by the New Jersey Center for Multimedia Research, by the National Center for Transportation and Industrial Productivity at the New Jersey Institute of Technology (NJIT), by the New Jersey Department of Transportation, by the New Jersey Commission of Science and Technology, as well as grants from the Sloane Foundation and the AT&T Foundation, and the NJIT SBR program.
     

    References

    ACKERMAN, M., Augmenting the organizational memory: a field study of Answer Garden. Proceedings of the ACM 1994 Conference on Computer Cooperative Work, ACM Press, October 1994, 243-252.

    AKSCYN, R., MCCRACKEN, D. AND YODER, E. KMS: a distributed hypermedia system for managing knowledge in organizations. Communications of the ACM 31(7) 1988, pages 820-835.

    ANDERSON, K. Providing automatic support for extra-application functionality. In [Ashman et al. 1996], 1996.

    ANDERSON, K. Integrating Open Hypermedia Systems with the World Wide Web. Hypertext'97 Proceedings, ACM Press, New York, NY, April 1997, 157-166.

    ANDERSON, K., TAYLOR, R. AND WHITEHEAD, E. Chimera: hypertext for heterogeneous software environments. Proceedings of the Fifth ACM Conference on Hypermedia Technologies, Edinburgh, Scotland, September 1994, 94-107.

    ASHMAN, H.L., BALASUBRAMANIAN, V., BIEBER, M. AND OINAS-KUKKONEN, H. (eds). Proceedings of the First International Workshop on Incorporating Hypertext Functionality Into Software Systems (HTF I), European Conference on Hypermedia Technologies (ECHT'94), Edinburgh, September 1994.

    ASHMAN, H.L., BALASUBRAMANIAN, V., BIEBER, M. AND OINAS-KUKKONEN, H. (eds). Proceedings of the Second International Workshop on Incorporating Hypertext Functionality Into Software Systems (HTF II), Hypertext '96 Conference, Bethesda, March 1996.

    ASHMAN, H.L., BALASUBRAMANIAN, V., BIEBER, M. AND OINAS-KUKKONEN, H. (eds). Proceedings of the Third International Workshop on Incorporating Hypertext Functionality Into Software Systems (HTF III), Hypertext '97 Conference, Southampton UK, April 1997.

    BANKER, R., ISAKOWITZ, T., KAUFFMAN, R., KUMAR, R. AND ZWEIG, D. Tools for managing repository objects. Software engineering economics and declining budgets, Geriner, T., Gulledge, T. and Hutzler, W. (eds.), Springer Verlag, 1993, 125-146.

    BAPAT, A., WÄSCH, J., ABERER, K. AND HAAKE, J. HyperStorM: an extensible object-oriented hypermedia engine. Hypertext'96 Proceedings, Washington, DC, March 1996, 203-214.

    BENNETT, K. Legacy systems: coping with success. IEEE Software, January 1995, 19-23.

    BHARGAVA, H. A logic model for model management. Ph. D. Dissertation, The Wharton School, University of Pennsylvania, Philadelphia, PA 19104, 1990.

    BHARGAVA, H. Using quiddities for detecting semantic conflicts in information systems. Journal of Organizational Computing 5(4), 1995, 379-400.

    BIEBER, M. Generalized hypertext in a knowledge-based DSS shell environment. Ph. D. Dissertation, The Wharton School, University of Pennsylvania, Philadelphia, PA 19104, 1990.

    BIEBER, M. Automating hypermedia for decision support. Hypermedia 4, No. 2 (1992) 83-110.

    BIEBER, M. On integrating hypermedia into decision support and other information systems. Decision Support Systems 14, 1995, 251-267.

    BIEBER, M. Developing a freight data executive MIS for New Jersey using hypertext. Technical Report CIS96-9, New Jersey Institute of Technology, Computer and Information Sciences Department, 1996.

    BIEBER, M. AND ISAKOWITZ, T. (eds.) Designing Hypermedia Applications. Special issue of the Communications of the ACM 38(8), 1995.

    BIEBER, M. AND ISAKOWITZ, T. Introduction to the special issue: hypermedia in information systems and organizations. Journal of Organizational Computing and Electronic Commerce 6(3), 1996, iii-vii.

    BIEBER, M., ISAKOWITZ, T. AND OLIVER, J. An analysis framework for information comprehension and access management. Proceedings of the Twenty-Eighth Annual Hawaii International Conference on System Sciences (Wailea, Maui; January 1995), IEEE Press, Washington, D.C., Volume IV, pages 603-613.

    BIEBER, M. AND KACMAR, C. Designing hypertext support for computational applications. Communications of the ACM 38(8), 1995, 99-107.

    BIEBER, M. AND KIMBROUGH, S. On generalizing the concept of hypertext. Management Information Systems Quarterly 16(1), 1992, 77-93.

    BIEBER, M. AND KIMBROUGH, S. On the logic of generalized hypertext. Decision Support Systems 11, 1994, 241-257.

    BIEBER, M. and VITALI, F. (1997). Toward support for hypermedia on the World Wide Web. IEEE Computer 30(1), 1997, 62-70.

    BIEBER, M., VITALI, F., ASHMAN, H., BALASUBRAMANIAN V., and OINAS-KUKKONEN, H. Fourth generation hypermedia: some missing links for the World Wide Web. forthcoming in the International Journal of Human Computer Studies.

    BIEBER, M. AND WAN, J. Backtracking in a multi-window hypertext environment. European Conference on Hypermedia Technologies'94 Proceedings (Edinburgh, Scotland; September 1994), ACM Press, 158-166.

    BOTAFOGO, R., RIVLIN, E. AND SHNEIDERMAN, B. Structural analysis of hypertext: identifying hierarchies and useful metrics. ACM Transactions on Information Systems 10(2), 1992, 142-180.

    BROWN, P. Assessing the quality of hypertext documents. Hypertext: Concepts, Systems and Applications, Proceedings of ECHT'90, Versailles, France, November 1990, Rizk, A. Streitz, N. and André, J. (eds), Cambridge Press, pages 1-12.

    BRUSILOVSKY, P., Methods and techniques of adaptive hypermedia. in User Modeling and User-Adapted Interaction 6, Kluwer, 1996, 87-129.

    BUFORD, J. Evaluating HyTime: an examination and implementation experience. Hypertext'96 Proceedings, Washington, DC, ACM Press, March 1996, 105-115.

    CALVI, L. AND DE BRA, P. Improving the usability of hypertext courseware through adaptive linking. Hypertext'97 Proceedings, ACM Press, New York, NY, April 1997, 224-225.

    CAMPBELL, B. AND GOODMAN, J. HAM: a general purpose hypertext abstract machine. Communications of the ACM 31(7), 1988, 856-861.

    CARLSSON, C. Fuzzy logic and hyperknowledge: a new effective paradigm for active DSS. Proceedings of the Thirtieth Annual Hawaii International Conference on System Sciences (Wailea, Maui; January 1997), IEEE Press, Washington, D.C., Volume V, pages 324-333.

    CARR, L., DE ROURE, D., HALL, W. AND HILL, G. The Distributed Link Service: a tool for publishers, authors and readers. Proceedings of the Web Revolution: Fourth International World Wide Web Conference, The Web Journal 1(1), O'Reilly and Associates, 1995.

    CATLIN, T., BUSH, P. AND YANKELOVICH, N., InterNote: extending a hypermedia framework to support annotative collaboration. Hypertext '89 Conference, ACM Press, 1989, 365-378.

    CHANG, A., HOLSAPPLE, C. AND WHINSTON, A. The hyperknowledge framework for decision support systems. Information Processing and Management, 1993.

    CHIU, C.M. Fusing the World Wide Web and the open hypermedia system technologies. Ph. D. dissertation, Rutgers University, Newark NJ 07102, 1997.

    CHIU, C. AND BIEBER, M. A generic dynamic-mapping wrapper for open hypertext system support of analytical applications. Hypertext'97 Proceedings, ACM Press, New York, NY, April 1997, 218-219.

    CONKLIN, E. J. Hypertext: a survey and introduction. IEEE Computer 20(9), 1987, 17-41.

    CROASDELL, D., PARADICE, D. AND COURTNEY, J. Using adaptive hypermedia to support organizational memory and learning. Proceedings of the Thirtieth Annual Hawaii International Conference on System Sciences (Wailea, Maui; January 1997), IEEE Press, Washington, D.C., Volume II, 281-289.

    DAVIS, H. (A) To embed or not to embed. Communications of the ACM. 38(8), 1995, pp 108-109.

    DAVIS, H. (B) Data integrity problems in an open hypermedia link service. Ph.D. Dissertation, University of Southampton, UK, 1995.

    DAVIS, H., HALL, W., I. HEATH, HILL, G. AND R. WILKINS. Towards an integrated information environment with open hypermedia systems. Proceedings of the ACM Conference on Hypertext (Milan, Nov. 1992) 181-190.

    DAVIS, H., KNIGHT, S. AND HALL, W. Light hypermedia link services: a study of third party application integration. Proceedings of the Fifth ACM Conference on Hypermedia Technologies, Edinburgh, Scotland, 1994, 158-166.

    DAVIS, H., LEWIS, A. AND RIZK, A. A draft proposal for a standard open hypermedia protocol. Proceedings of the Second Workshop on Open Hypermedia Systems, in [Wiil and Demeyer 1996].

    DELISLE, N. AND SCHWARTZ, M. Neptune: a hypertext system for CAD applications. Proceedings of the 1986 ACM SIGMOD Conference, May 1986, pages 132-143.

    DEROSE, S. Expanding the notion of links. Hypertext '89 Proceedings, Pittsburgh, PA, November 1989, 249-257.

    DIAZ, A. AND ISAKOWITZ, T. RM Case: Computer-aided support for hypermedia design and development. in [Fraïssé et al. 1995], 3-16.

    DIAZ, A., ISAKOWITZ, T., MAIORANA, V. AND GILABERT, G. RMC: a tool to design WWW applications. Proceedings of the Fourth International World Wide Web Conference, Boston, December 1995, http://www.w3.org/WWW4.

    ENGELBART, D. Knowledge domain interoperability and an open hyperdocument system. Proceedings of the Conference on Computer-Supported Cooperative Work, ACM Press, New York, 1990,143-156.

    FOUNTAIN, A., HALL, W., HEATH, I. AND DAVIS, H. MICROCOSM: An open model for hypermedia with dynamic linking. Proceedings of European Conference on Hypertext (ECHT) '90 (Cambridge University Press, Versailles, Nov. 1990) 298-311.

    FRAïSSÉ, S., GARZOTTO, F., ISAKOWITZ, T., NANARD, J. AND NANARD, M. (eds.), International Workshop on Hypermedia Design'95 Proceedings, Montpellier (June 1-2, 1995).

    GAMMA, E., HELM, R., JOHNSON, R. AND VLISSIDES, J. Design patterns: elements of reusable object-oriented software, Addison-Wesley, 1995.

    GARRIDO, A. AND ROSSI, G. A framework for extending object-oriented applications with hypermedia functionality. The New Review of Hypermedia and Multimedia 2, 1996, 25-41.

    GARZOTTO, F., MAINETTI, L. AND PAOLINI, P. Hypertext design, analysis, and evaluation issues. Communications of the ACM 38(8), August 1995, 74-86.

    GARZOTTO, F., MAINETTI, L. AND PAOLINI, P. (A) Navigation patterns in hypermedia data bases. Journal of Organizational Computing and Electronic Commerce 6(3), 1996, 211-237.

    GARZOTTO, F., MAINETTI, L. AND PAOLINI, P. (B) Information reuse in hypermedia applications. Hypertext'96 Proceedings, Washington, DC, ACM Press, March 1996, 93-104.

    GARZOTTO, F., PAOLINI, P. AND SCHWABE, D. HDM - a model-based approach to hypertext application design. ACM Transactions on Information Systems 11, No. 1 (1993) 1-26.

    GRØNBÆK, K., HEM, J., MADSEN, O. AND SLOTH, L. Cooperative hypermedia systems: a Dexter-based architecture. Communications of the ACM 37(2), 1994, 64-74.

    GRØNBÆK, K. AND TRIGG, R. Design issues for a Dexter-based hypermedia system. Communications of the ACM 37(2), 1994, 40-49.

    GRØNBÆK, K. AND WIIL, U. Towards a reference architecture for open hypermedia. Proceedings of the Third Workshop on Open Hypermedia Systems, in [Wiil 1997].

    HALASZ, F. Reflections on NoteCards: seven issues for the next generation of hypermedia systems. Communications of the ACM 31(7), 1988, 836-855.

    HALASZ, F. AND SCHWARTZ, M. The Dexter hypertext reference model. (edited by Grønbæk, K. and Trigg, R.), Communications of the ACM 37(2), 1994, 30-39.

    HARDMAN, L., BULTERMAN, D. AND VAN ROSSUM, G. The Amsterdam hypermedia model: adding time and context to the Dexter model. Communications of the ACM 37(2), 1994, 50-62.

    HILL, G. AND HALL, W. Extending the Microcosm model to a distributed environment. European Conference on Hypermedia Technologies'94 Proceedings (Edinburgh, Scotland; September 1994), ACM Press, 32-40.

    HIRATA, K., HARA, Y., SHIBATA, N. AND HIRABAYASHI, F. Media-based navigation for hypermedia systems. Hypertext'93 Proceedings, Seattle, November 1993, 159-173.

    HIRATA, K., HARA, Y., TAKANO, H., AND KAWASAKI, S., Content-oriented Integration in Hypermedia Systems. Hypertext'96 Proceedings. Washington, DC, March 1996, 11-21.

    INTERNATIONAL STANDARDS ORGANIZATION IEC IS 10744 Hypermedia/time-based document structuring language (HyTime). August 1992.

    ISAKOWITZ, T., Hypermedia in information systems and organizations: a research agenda. Proceedings of Twenty-Sixth Annual Hawaii International Conference on System Science (HICSS) (Maui, Jan. 1993), Volume III, 361-369.

    ISAKOWITZ, T., KAMIS, A. AND KOUFARIS, M. Extending the capabilities of RMM: Russian dolls and hypertext. Proceedings of the Thirtieth Annual Hawaii International Conference on System Sciences (Wailea, Maui; January 1997), IEEE Press, Washington, D.C., Volume VI, pages 177-186.

    ISAKOWITZ, T. AND STOHR, E. Hypertext-based relationship management for DSS. New York University Information Systems Department Working Paper IS-92-22 (New York, NY 10012, July 1992).

    ISAKOWITZ, T., STOHR, E. AND BALASUBRAMANIAN, P. RMM: A methodology for structuring hypermedia design. Communications of the ACM 38(8), August 1995, 34-44.

    KACMAR, C., Supporting hypermedia services in the user interface. Hypermedia 5(2), 1993, 85-101.

    KACMAR, C. A process approach for providing hypermedia services to existing, non-hypermedia applications. Electronic Publishing: Organization, Dissemination, and Distribution 8(1), March 1995, 31-48.

    KACMAR, C. AND LEGGETT, J. PROXHY: a process-oriented extensible hypertext architecture. ACM Transactions on Information Systems 9, No. 4 (1991) 399-419.

    KIMBROUGH, S. On the reduction of genetics to molecular biology. Philosophy of Science 46(3), 1979, 389-406.

    KIMBROUGH, S., PRITCHETT, C., BIEBER, M. AND BHARGAVA, H. The Coast Guard's KSS project. Interfaces 20, No. 6 (1990) 5-16.

    KIMBROUGH, S. AND VACHULA, W. A propædeutic on logic for object-oriented systems analysis and design. Proceedings of the Twenty-Ninth Annual Hawaii International Conference on System Sciences (Wailea, Maui; January 1996), IEEE Press, Washington, D.C., Volume II, 20-29.

    JOHNSON, S. Quality control for hypertext construction. Communications of the ACM 38(8), 1995, 87.

    LANDOW, G. Popular fallacies about hypertext. in Jonassen, D. and Mandl, H. (eds.) Designing hypermedia for learning, Series F. Computer and Systems Sciences 67, Springer-Verlag, 1990, 39-59.

    LANGE, D. Object-Oriented Hypermodeling of Hypertext Supported Information Systems. Journal of Organizational Computing and Electronic Commerce 6(3), 1996, 269-293.

    LEE, Y. K., YOO, S-J., YOON, K. AND BERRA, P. Querying structured hyperdocuments. Proceedings of the Twenty-Ninth Annual Hawaii International Conference on System Sciences (Wailea, Maui; January 1996), IEEE Press, Washington, D.C., Volume II, pages 155-164.

    LEGGETT, J. AND SCHNASE, J. Viewing Dexter with open eyes. Communications of the ACM 37(2), 1994, 77-86.

    LEWIS, A. DAVIS, H., GRIFFITHS, S., HALL, W. AND WILKINS, R. Media-based navigation with generic links. Hypertext '96 Proceedings, ACM Press, 1996, 215-223.

    LITTLEFORD, A. Artificial intelligence and hypermedia. in: Berk, E. and Devlin, J. (eds.) Hypertext/Hypermedia Handbook (Intertext Publications/McGraw-Hill Publishing Co., Inc., New York, 1991), 357-378.

    LUCARELLA, D. AND ZANZI, A. A visual retrieval environment for hypermedia information systems. ACM Transactions on Information Systems 14(1), 3-29.

    MAIOLI, C., PENZO, W., SOLA, S. AND VITALI, F. Using a reference model for information system compatibility: an experience. Proceedings of the Twenty-Seventh Hawaii International Conference on System Sciences, January 1994, Maui, Hawaii, Volume III, 376-385.

    MAIOLI, C., SOLA, S. AND VITALI, F. Wide-area distribution issues in hypertext systems. ACM SIGDOC'93 Conference Proceedings, ACM Press, New York, 1993.

    MALCOLM. K., POLTROCK, S. AND SCHULER, D. Industrial strength hypermedia: requirements for a large engineering enterprise. Hypertext `91 Proceedings, December 1991, 13-24.

    MAO, J., BENBASAT , I. AND DHALIWAL, J. Enhancing explanations in knowledge-based systems with hypertext. Journal of Organizational Computing and Electronic Commerce 6(3), 1996, 211-237.

    MARSHALL, C. AND SHIPMAN III, F. Spatial hypertext: designing for change. Communications of the ACM 38(8), 1995, 88-97.

    MARSHALL, C., SHIPMAN III, F. AND COOMBS, J. VIKI: spatial hypertext supporting emergent structure. European Conference on Hypermedia Technologies'94 Proceedings (Edinburgh, Scotland; September 1994), ACM Press, 13-23.

    MARSHALL, C., SHIPMAN III, F. AND MCCALL, R. Making large-scale information resources serve communities practice. Journal of Management Information Systems 11(4), 1995, 65-86.

    MARSHALL, C. AND IRISH, P. Guided tours and on-line presentations: how authors make existing hypertext intelligible for readers. Hypertext '89 Proceedings (Pittsburgh, Nov. 1989) 15-42.

    MEYROWITZ, N. The missing link: why we're all doing hypertext wrong. in The society of text: hypertext, hypermedia, and the social construction of information, Barrett, E. (ed.), MIT Press, Cambridge, MA, 1989, 107-114.

    MINCH, R. AND GREEN, G. An investigation of hypermedia support for problem decomposition. Journal of Organizational Computing and Electronic Commerce 6(3), 1996, 295-323.

    NANARD, J. AND NANARD, M. (A) Hypertext design environments and the hypertext design process. Communications of the ACM 38(8), 1995, 49-56.

    NANARD, J. AND NANARD, M. (B) Adding macroscopic semantics to anchors in knowledge-based hypertext. International Journal of Human-Computer Studies 43, 1995, 363-382.

    NEWCOMB, S., KIPP, N. AND NEWCOMB, V. The 'HyTime' hypermedia/time-based document structuring language. Communications of the ACM 34(11), 1991, 67-83.

    NIELSEN, J. The art of navigating through hypertext. Communications of the ACM 33(3), March 1990, 296-310.

    NIELSEN, J. Multimedia and hypertext: the Internet and beyond. AP Professional, 1995.

    NOLL, J. AND SCACCHI, W. Repository support for the virtual software enterprise. In [Ashman et al. 1996].

    OINAS-KUKKONEN, H. Debate Browser - an argumentation tool for MetaEdit+ Environment. Proceedings of the Seventh European Workshop on Next Generation of CASE Tools (NGCT '96), Crete, Greece, 1996, 77-86.

    OINAS-KUKKONEN, H. Embedding hypermedia into information systems. Proceedings of Thirtieth Annual Hawaii International Conference on System Science, IEEE Computer Society Press, Los Alamitos, 1997, Volume VI 187-196.

    ØSTERBYE, K. AND WIIL, U. The Flag Taxonomy of open hypermedia systems. Hypertext'96 Proceedings, ACM Press, New York, 1996, 129-139.

    PARUNAK, H. (A) Ordering the information graph. in Hypertext/hypermedia handbook, Berk, Emily and Joseph Devlin (eds.), Intertext Publications/McGraw-Hill Publishing Co., Inc., New York, 1991, 299-325.

    PARUNAK, H. (B) Toward Industrial Strength Hypermedia. in Hypertext/hypermedia handbook, Berk, Emily and Joseph Devlin (eds.), Intertext Publications/McGraw-Hill Publishing Co., Inc., New York, 1991, 381-395.

    PEARL, A. Sun's Link Service: a protocol for open linking. Hypertext '89 Proceedings (Pittsburgh, Nov. 1989) 137-146.

    PUTTRESS, J. AND GUIMARAES, N. The toolkit approach to hypermedia. Hypertext: Concepts, Systems and Applications, Proceedings of ECHT'90, Versailles, November 1990, Rizk, A., Streitz, N. and André, J. (eds.), Cambridge University Press, 25-37.

    RANA, A. AND BIEBER, M. Towards a collaborative hypermedia education framework. Proceedings of Thirtieth Annual Hawaii International Conference on System Science (HICSS) (Maui, Jan. 1997), II 610-619.

    RAO, U. AND TUROFF, M., Hypertext functionality: a theoretical framework. International Journal of Human-Computer Interaction 2(4), 1990, 333-358.

    RIZK, A. AND SAUTER, L. Multicard: an open hypermedia system. Proceedings of the ACM Conference on Hypertext (Milan, Nov. 1992) 4-10.

    RIVLIN, E., BOTAFOGO, R. AND SHNEIDERMAN, B. Navigating in hyperspace: designing a structure-based toolbox. Communications of the ACM 37(2), 1994, 87-108.

    ROSSI, G., GARRIDO, A. AND CARVALHO, S. Design patterns for object-oriented hypermedia applications. In: Pattern languages of programs II. Vlissides, J., Coplien, J. and Kerth, N. (eds.) Addison-Wesley, 1996, pp. 177-191.

    ROSSI, G., SCHWABE, D. AND GARRIDO, A. Design reuse in hypermedia application development. Hypertext'97 Proceedings, ACM Press, New York, 1997, 57-66.

    SCACCHI, W. On the power of domain-specific hypertext environments. Journal of the American Society for Information Science 40(3), 1989, pages 183-191.

    SHERMAN, M., HANSEN, W., MCINERNY, M., AND NEUENDORFER, T. Building hypertext on a multimedia toolkit: an overview of the Andrew toolkit hypermedia facilities. Proceedings of European Conference on Hypertext (ECHT) '90 (Cambridge University Press, Versailles, Nov. 1990) 13-24.

    SCHÜTT, H. AND STREITZ, N., HyperBase: a hypermedia engine based on a relational database management system. In: Rizk, A., Streitz, N. and André, J. (eds.) Hypertext: Concepts, Systems and Applications, Proceedings of ECHT'90, Versailles, November 1990. Cambridge: Cambridge University Press, 95-108.

    SCHWABE, D., ROSSI, G. AND BARBOSA. S. Systematic hypermedia application design with OOHDM. Hypertext'96 Proceedings, March 16-20, 1996, 116-128.

    STOTTS, P. D. AND FURUTA, R. Petri-net-based hypertext: document structure with browsing semantics. ACM Transactions on Information Systems 7(1), 1989. 3-29.

    TANIK, M., YEH, R., BIEBER, M., KURFESS, F., LIU, Q., MCHUGH, J., RANA, A., ROSSAK, W. AND NG, P. Issues and architecture for electronic enterprise engineering (EEE). Proceedings of the Second World Conference on Integrated Design and Process Technology, Society for Design and Process Science, (Austin, Dec. 1996), Volume 2, 57-62. [http://www.cs.njit.edu/~eee/pubs/sdps96.html]

    TERVONEN, I., KEROLA, P. AND OINAS-KUKKONEN, H. An organizational memory for quality-based software design and inspection: a collaborative multiview approach with hyperlinking capabilities. Proceedings of the Thirtieth Annual Hawaii International Conference on System Sciences (Wailea, Maui; January 1997), IEEE Press, Washington, D.C., Volume II, 290-299.

    THÜRING, M., HANNEMANN, J. AND HAAKE, J. Hypermedia and cognition: designing for comprehension. Communications of the ACM 38(8), 1995, 57-69.

    TRIGG, R. AND WEISER, M. TextNet: A network-based approach to text handling. ACM Transactions on Office Information Systems 4, No. 1 (1986) 1-23.

    UTTING, K. AND YANKELOVICH, N. Context and orientation in hypermedia networks. ACM Transactions on Information Systems 7, No. 1 (1989) 58-84.

    VANHARANTA, H., KÄKÖLÄ, T. AND BACK, B. Validity and utility of a hyperknowledge-based financial benchmarking system. Proceedings of the Twenty-Eighth Hawaii International Conference on System Sciences, January 1995, Maui, Hawaii, Volume III, 221-230.

    WAN, J. Integrating hypertext into information systems through dynamic linking. Ph. D. dissertation, New Jersey Institute of Technology, Institute for Integrated Systems Research, Newark NJ 07102, 1996.

    WAN, J. AND BIEBER, M. GHMI: a general hypertext data model supporting integration of hypertext and information systems. Proceedings of the Twenty-Ninth Annual Hawaii International Conference on System Sciences (Wailea, Maui; January 1996), IEEE Press, Washington, D.C., Volume II, pages 47-56.

    WAN, J. AND BIEBER, M. Providing relational database management systems with hypertext. Proceedings of the Thirtieth Annual Hawaii International Conference on System Sciences (Wailea, Maui; January 1997), IEEE Press, Washington, D.C., Volume VI, 160-166.

    WANG, W. AND RADA, R. experiences with semantic net based hypermedia. International Journal of Human-Computer Studies, 43, 419-439, 1995.

    WHITEHEAD, E. J. An architectural model for application integration in open hypermedia environments. Hypertext'97 Proceedings, ACM Press, New York, 1997, 1-12.

    WIIL, U. (ed.). Proceedings of the 3rd Workshop on Open Hypermedia Systems. CIT Scientific Report #SR-97-01, 1997, The Danish National Centre for IT Research, Forskerparken, Gustav Wieds Vej 10, 8000 Aarhus C, Denmark [on-line] http://www.cit.dk/

    WIIL, U. AND DEMEYER, S. (eds). Proceedings of the 2nd Workshop on Open Hypermedia Systems. at Hypertext'96, Washington, March 16-17, 1996. UCI-ICS Technical Report 96-10, Department of Information and Computer Science, University of California, Irvine, CA 92717-3425.

    WIIL, U. AND LEGGETT, J. The HyperDisco approach to open hypermedia systems. Hypertext'96 Proceedings, ACM Press, New York, 1996, 140-148.

    Footnotes

    F1. The term hypermedia nominally applies hypertext concepts to multiple media [Hardman et al. 1994; Hirata et al. 1993, 1996; Lewis et al. 1996; Lucarella and Zanzi 1996]. Hypermedia researchers usually use the terms interchangeably.

    F2. We use the term "dynamically" to mean calculated on the fly at run time. This has nothing to do with other meanings, including the notion of active media such as video or animation.

    F3. Those interested in writing processes for software reuse may wish to reference the following research utilizing hypermedia techniques for software engineering: [Banker et al. 1993; Rossi et al. 1997]; and for hypermedia applications [Garzotto et al. 1996b]. Garzotto et al. [1995] also give some warnings for reusing hypermedia presentations.

    F4. We note that in the first version of the prototype, the DHTE's standard UIS message format is very close to Microcosm's message format. This would require other UIS wrappers to convert messages between this format and their native UISs' message format. Future releases of the DHTE will use a more generic standard message format.