The composite information units of hypermedia: Hyperdocuments and database diagrams

Hypermedia consists, roughly speaking, of a network of links which connect nodes. A contained network constitutes a hyperdocument.

Nodes and links may also be visualised in a database diagram, either as an entire hyperdocument or as parts of a hyperdocument.

This section of the lesson will examine the hyperdocument and the database diagram.

The composite information units of hypermedia

This section examines what we have chosen to call the composite information units of hypermedia, which comprise hyperdocuments and database diagrams. These are hypermedia units which are constructed of the basic hypermedia units, links and nodes. A composite information unit is a conceptual term, and a composite information unit does not have to constructed by an existing (basic) information unit. A database diagram (at times simply referred to as a diagram) may be constructed manually by its author. Furthermore, a diagram does not need to be represented as a separate unit at all, but it may be dynamically generated by the system according to the user's whereabouts in the diagram.

A hyperdocument is a collection of nodes and links. It may often deal with a particular topic, and may be compared to an article in an encyclopaedia or a book examining a limited field. A hyperdocument will often have a particular structure, which is normally hierarchical, but other structures may also occur.

A database diagram is a graphic overview displaying the structure of all or parts of a hyperdocument. The nodes are usually displayed as boxes or by link icons in the diagram, and the links are usually indicated by lines between the nodes. A hyperdocument may also contain a database diagram. A reader may generally move in the hyperdocument by means of this database diagram. In some systems the author may perform changes directly in the diagram. For example, some systems permit nodes to be deleted or created directly from the diagram.

Hyperdocuments

Hypermedia uses the document metaphor to describe a collection of nodes, links and diagrams which typically cover a particular, delimited topic. A hyperdocument may be part of a larger hypermedia base, or the document may in itself be a hypermedia base. However, deviating from the definition above, some systems use the term document about nodes, for example Intermedia (Yankelovich, Haan, Meyrowitz, Drucker 88). In Xanadu (Nelson 90) the term "document" is applied to any separate unit, for example an available link to a topic, a node, or a collection of several objects. The latter may also be a document according to the definition above.

Beyond the various hypermedia units (links, nodes and diagrams) stored in the hyperdocument, so-called context information is often also stored. This information describes various possible states a hypermedia system may be in. A state may exist based on different conditions. One condition may be which information must be visible (filtering) to the user in the selected document. Other conditions may be setting which node or nodes should be open on start-up, and which paths or routes should be accessible beyond the existing links. The system might also recall what the user did when he or she last used the system.

Frank Halasz mentions so-called composites (Halasz 88, p. 843-845). These are collections of nodes and links within a document which may be managed as a separate entity. This permits operations on the composite as a unit. This is the opposite of a regular document, where operations must be carried out on each of its constituent units. A composite may thus be considered a sub-document. However, it is difficult to establish any large conceptual difference between a document and a composite. We will thus consider documents and composites as equals.

Document representation

As stated in the introduction to this section, "composite hypermedia unit" is a conceptual term. This means that the conceptual unit hyperdocument may be represented in the information model in various ways in a hypersystem.

One option is to let the document constitute a separate entity, where nodes and links are represented. Node and link representation may be done in two ways: directly or indirectly. In the direct method the document contains a set of the nodes and links which are part of the document. The links and nodes will thus be hyperdocument components. This applies to NoteCards (Halasz, Moran, Trigg 87). The figure below shows the principle for the direct method of document representation.

In the indirect method the document contains a description of characteristics, in the form of attribute values of links and nodes which are included in the document. A special variation of this is used in Intermedia, where the document entity contains information about the information objects which are to be used as nodes and the placement of links in the nodes. When the document is opened the hypersystem constructs nodes based on the information objects and the link information present in the document. This enables using the same information objects as a starting point for a number of nodes in different documents. This makes it possible to present the same information in a number of variations. See section 2.2.4 for more on node representation in the information model.

Another option is to have no explicit representation of the document at all. By opening an available node and following links from this, it is possible to move around in a set of nodes. Conceptually such a set of nodes must be considered a document. The requirement is that the user has direct access to a node, which will then function as an access node to the hyperdocument. If the system has one-way links, the document will appear as a contained unit for the user. When two-way links are used, the user may move freely anywhere in the document, thus perhaps no longer sensing that it is a separate document. KMS (Akscyn, McCracken, Yoder 88) utilises this type of organising the hyperbase into documents. Figure 3.2 shows the principle of this method.

The third option is to allow nodes, links and documents to be separate entities in the same hyperbase. This is the most flexible way of representing documents. In an ideal system a document may consist of an open quantity of links and nodes. Different documents may wholly or partly consist of the same links and nodes, i.e. a mathematical union. A node may therefore be linked to another node or to another document. Links, nodes and documents may all be handled as separate entities on which operations may be performed. This representation is used by Xanadu, which is probably the closest relative of Halasz' composites.


The structure of hyperdocuments

The structure of a hyperdocument is the overarching organisation of the document nodes. By overarching organisation we mean the primary order in which information is presented as chosen by the author. The structure is maintained by organisation links connecting the nodes.

An important issue much discussed over the years has been which structures should be permissible in hypermedia, and/or whether hypermedia systems should offer a fixed structure at all. Theodor Holm Nelson and Douglas Engelbart disagreed on this, even though both visualised the great possibilities hypermedia would be able to offer as a tool "to extend the human intellect" (Engelbart) and as a "thinkertoy" (Nelson). Engelbart believes that hypermedia should use tree structures, while Nelson feels that the user's own mental model of the information should be implemented in the hyperdocument.

DeRose (DeRose 89, p. 249) points out that traditional documents usually have a number of structures. They have two main structures, the first being the purely linguistic structure of regular language, with chapters, sections etc. The second structure is the physical structure of pagination, lines etc. DeRose believes that hypermedia should preferably be able to represent several structures in a hypermedia document. Jeff Conclin shares this belief. He draws special attention to the fact that it must be possible to present the information in a number of parallel tree structures (Conclin 87, p. 35).

Most hypersystems require or at least encourage authors to structure hyperdocuments hierarchically. As there is also a need for other structures, in the following section we will examine different ways of structuring hyperdocuments.

One aspect of hyperdocument structuring is that it is rather the specialised applications with hyper-functionality which have used structures other than the tree structure. One reason for this may be that these specialised applications may have been designed by specialists, while typical hypersystems are intended as authoring tools for normal users. Therefore document structuring has never been all that much of an issue when discussing hypermedia systems. Therefore, one of the sources for the following sections is Apple's Design Guidelines (Apple 89), which are intended for designers using HyperCard.

Free document structuring

Theodor Holm Nelson argues that the user's own concept of the information must be modelled. Nelson states that a hypermedia system must be able to represent any structure the user wants. He writes the following on the requirements for a hypermedia storage system (Nelson 90, p2/26):

" ... It is therefore the representation of this structure -- of whatever structure the user may be concerned with -- that should concern us. But the things you work on can be staggeringly complex. So we must create a general system of representation that will automatically hold and present any possible structures you want to work on."

Such an optional document structuring may manifest itself in various ways. It may either result in a more or less haphazard network structure, or the author might construct some kind of systematic structure.

A random network is not linked according to any system. The user constructs new links as his ideas develop, without having to adapt to any pre-defined structure. As this type of network does not possess a fixed structure, the reader may become confused. Understanding the architecture might be hard, thus making it difficult to move around in the document (Apple 89).

Even if a hypersystem permits an author the use of a free structure, s/he may nevertheless choose to employ a more systematic structure. Nelson believes that templates should be used wherever possible. The author then defines a template, e.g. a node template, which will contain a number of links to maintain the structure of the document.

Tree structure

With some exceptions, the tree structure has been the most common structure in existing systems. A drawback of organising information in this way is that it compels the author to structure his topic into a hierarchy. For readers, however, two special advantages are connected to tree structures. Firstly, navigating in a tree structure is easier. A reader may move back to the parent node, or on to one of the offspring nodes (in addition to any reference links). Moreover, it is easy for a user to visualise a tree structure, thus making it easier to orient him/herself in the document.

The figure below shows three possible tree structures (taken from Apple (Apple 89, p. 26-28). The topmost structure is often used in hypermedia systems, for example KMS. Each node may have a variable number of offspring nodes, and a parent node. The next structure also allows moving between sibling nodes. The final structure is more special, as no branch of the tree is permitted to branch further. This structure may be compared to that of a book, where a user may jump to any chapter from the table of contents.


Linear structures

Linear structures consist of a series of nodes. A reader may thus follow links in two directions only, forward to the next node or back to the previous node. Conceptually, this structure has characteristics in common with guided tours, see section 3.1 in lesson 7. This structuring method lends itself well where there is no requirement that a user can move around in the document. The drawback of this method is that the user cannot move in the document at will. Typical applications are demonstrations which will be familiar to many from presentation software such as Microsoft Powerpoint.

This method could also be used to transfer documents from text to hypertext, as the original document structure could be retained. Albeit this is a simple structuring method, it retains much of the strength of hypertext because it uses reference links.

Another possible application area is in instruction of training documents. Normally teaching material builds on other material, and a linear structure forces the reader to study the first node before proceeding to the next. The reader must see the elementary material first.


Single nodes

There may be a frequent need to present more information in a single window than usually will be possible in a node. This applies to documents where the reader logically feels that several activities are going on in a node. The reader may in such cases feel that navigation does not exist and/or is illogical (Apple 89, p. 30).

Figure 3.5 shows how such a node would look like. The information typically comes from specialised hardware such as a CD-ROM player or a videodisk player, or from a database, e.g. a picture database. A reader may navigate in the information using typical video-player controls. Even if the reader is moving around in the information, he will at all times feel that he is in the same node. Of course, it may be argued whether this is a node, or if each new picture corresponds to an independent node, as each node typically is meant to represent a contained piece of information. Conversely, it may be claimed that such pictures may often come from static media from hardware outside the hyperbase itself. Nevertheless, the reader (and frequently also the author) will perceive this information as an integral part of the hyperbase. Conceptually, therefore, single nodes may also be considered as several nodes, thus having points of similarity with sequential documents.

A special application for the single node is the filter stack (Apple 89, p. 31). Here the node is used for database queries for occurrences that satisfy a number of conditions. Such nodes are often used for species classification, for example determining plant species (cf. an early software product from Datasekretariatet [The Data Secretariat] called Dataflora). Each condition, such as leaf shape or flower colour, is entered in a field on the screen, and described by means of information objects. The reader may then change the query conditions by moving in the database using mouse clicks. When a number of conditions have been selected the query can be initiated. Figure 3.6 shows a node for determining mushroom species.


Links through hyperdocuments

If the hypersystem permits links across documents, a problem will arise: should the reference node also be visible to the user? If a node in one document includes a link to a node in another document, there are two ways of understanding the reference node document. Figure 3.7 below shows this. Moreover, the information model may or may not permit links across documents. This is not feasible in NoteCards, for example.



Database diagrams

A database diagram is a graphic presentation of links and other information units in the entire hyperbase or parts of it, commonly within one single hyperdocument. The links are usually rendered as lines between the other information units, which are usually rendered as "boxes" or as link icons. The other information units represented in the diagram are most frequently nodes, but also other information units may be represented, for example hyperdocuments and other database diagrams.

In some systems, such as Neptune and gIBIS, diagrams are purely tools, employed by the user when needed, for example the spell-checker tool in a word processor. In other systems, such as NoteCards and Intermedia, the diagram is also a separate information unit, which is represented in the information model along with the other information units. In the latter case the author may therefore construct diagrams which are specialised for different uses. Furthermore, it is frequently possible to perform the same operations on the diagram, as on the other information units, for example deletions and additions.

Database diagrams have two main purposes: speedy navigation among hyperbase nodes and simple hyperbase orientation. A user may usually move directly in the diagram by pointing at the node he or she wants to move to. Conceptually, database diagram navigation has similarities with link following. However, there is one essential difference. When following links, the user may only move to those links which are connected to the open node. By navigating in a database diagram, the user may move to any node in the diagram. The advantage is that the user can move across larger areas each time.

Examining the database diagram enhances a user's understanding of hyperbase structure, thus allowing improved navigation. If a user has gone to the wrong place when navigating due to orientation problems, he or she can move directly to the right node. In addition to diagrams there are also other mechanisms which may reduce orientation problems.

Database diagrams may permit users to enter editing mode on links and nodes, e.g. in NoteCards. Some systems also allow user operations on the hyperbase directly in the diagram, without having to go to the individual link or node. In NoteCards an author may delete and construct links and nodes from the database diagram.

The link indications in diagrams may also include further information than pure structure information. Both NoteCards and gIBIS show the link type in their link indication in their database diagrams. NoteCards uses various types of dotted lines, while gIBIS applies various colours to indicate different link types.

Nodes may be visualised in different ways. A common method is to render the node as a box with the node name in it. This applies to the Neptune and NoteCards systems. gIBIS uses various icons to indicate node types. The node name is placed underneath each icon. Intermedia presents nodes through the individual node icon with the node name underneath. Intermedia employs different icons to indicate the type of information object contained by the node.

Generating database diagrams

There are three usual methods for generating diagrams: manually constructed diagrams, diagrams generated by the system and stored statically, and dynamically generated diagrams. Manually constructed diagrams are made by the author. Typically such a diagram is a regular node with the chart represented as meta-graphics drawn by the author.

The second method is to have the hypersystem generate the diagram and then store this as a separate information unit. NoteCards uses this method. A database diagram may then be treated as a regular node. There may be problems if the hyperbase is modified, as the diagram will then no longer be valid. This can be overcome if hyperbase modifications automatically trigger updates of the database diagrams. NoteCards has this feature. If such a feature is not included in a system, an author should at least be notified about hyperbase modifications, so he or she can generate a new diagram.

The third method is generating diagrams dynamically each time the application is executed. Such diagrams will therefore always show the actual structure of the hyperdocument, including the latest hyperbase changes. Intermedia generates a new local database diagram - see the following section on local diagrams - as the reader moves around in the document. Another example of dynamically generated diagrams is found in Neptune, where diagrams are constructed on the user's request.

However, generating a diagram of a certain size requires much processing power. The time required to locate all connections among network nodes in a general network is proportional at N2 + KN where N is the number of network nodes and K is the number of edges (Utting, Yankelovich 89, p. 74). If we assume that this can be transferred to apply to hypermedia, where network nodes correspond to nodes and network edges correspond to links, this constitutes the basis for a viable time estimate.

Furthermore, database diagrams do not lend themselves to distributed environments, where nodes and links may be distributed across a number of servers, as it will take time to search through several servers.

In a distributed environment database diagrams require much extra work, where nodes and links may be distributed across a number of servers, as it will take time to search through several servers. One of the systems offering diagrams in distributed environments is HyperWave.

Specialising diagrams

No special reservations have been made in the preceding sections as to which parts of the hyperbase are included in the diagram. However, a hyperbase comprising a number of larger hyperdocuments may become quite big, making it inappropriate to put all information units in the hyperbase directly in one, single (global) diagram. If diagrams become too large, they become hard to read, and hence lose their value for the user. This has been documented by Utting, Yankelovich (Utting, Yankelovich 89, p. 62) and by Conclin (Conclin 87, p.39).

It has therefore become the norm to limit the amount of information to enhance diagram readability. In the following sections we will examine some of the techniques used, and also some which have been proposed to limit the amount of information in diagrams.

Limiting information unit attributes

Limiting attribute values and information content for those information units which are going into a diagram will reduce diagram size. Some hypersystems offer the option of restricting the information amount via query and filtering processes. Using natural language syntax this might, for example, be "Give me all information units authored by members of group 5, after 14 November, except those written by member NN". In Neptune this can be done directly by the reader. Database diagrams have separate fields where the user may set valid ranges for information unit attributes (nodes and links) which are to be included. One drawback of this system - that the reader does not always know what he is looking for - has been pointed out by Utting, Yankelovich (Utting, Yankelovich 89, p. 66-68).

In some hyperbases which treat database diagrams as separate information units, the author may construct the diagram. This ensures that the user is provided with a diagram containing the relevant information. However, this is not possible in systems such as Neptune and gIBIS, which treat diagrams more as special tools.

If the system offers filtering of some kind, this may also be reflected in the diagram. Filtering implies restricting attribute values and information content in the part of the information base which is visible to the user. Only those information objects left after filtering will be included in the diagram. There will be more about filtering in the next lesson.

3.3.2.2 Fisheye database diagrams

Fisheye diagrams have been proposed by Furnas (Furnas 86). A fisheye diagram offers a higher detail level for those information units which are closest to the user's position at any time. The closest area may offer full details including all links and connected nodes. Farther away, only organisation links may be shown, while in the most distant parts landmark nodes and links to these will be sufficient.

Local diagrams

Local diagrams show only sibling nodes and their links. Such diagrams are used in Intermedia, as a consequence of the global database diagram problems. This diagram tracks user movements, always letting the user know where he or she may move.

Utting and Yankelovich (Utting, Yankelovich 89) have extended the local diagram. They have emphasised that a diagram only addresses some of the orientation problems users encounter. A user needs additional information about where he or she just came from in order to avoid going in circles. Moreover, the user must know when he or she has visited a location. The outcome of this is an improved diagram for Intermedia, shown on the next page. Using history and time tracking for the most recent visit are examined further in the following lesson.

Multi-level diagrams

When utilising a number of diagrams, each diagram may offer a different level of detail for the same information. Users may use a coarser detail level for a better overview. Coarse diagrams often display only structure, not giving information such as the name and types of the information units which are represented in the database diagram. Using the coarse diagram it is possible to come up with rough estimates as to the extent of the available information (Utting, Yankelovich 89, page 66). One possible drawback to multi-level diagrams is that these in themselves may cause orientation problems (Nielsen 90, p. 300).

gIBIS uses two-level diagrams, one coarse, one fine. Database diagrams in gIBIS consist of a window with a large, high-detailed diagram, and a smaller, coarse-detailed diagram. The smaller diagram offers an overview of the major part (all) of the base. The big diagram offers an overview of the user's present whereabouts. A user may furthermore navigate to another location by clicking on it.