Navigational Techniques and Tools

TBL
There are a number of ways of accessing the data one is looking for. Navigational access (i.e., following links) is the essence of hypertext, but this can be enhanced with a number of facilities to make life more efficient and less confusing.

Defined structure

It is sometimes nice for a reader to be able to reference a document structure built specifically to enhance his understanding, by the document author. This is especially important when the structure is part of the information the author wishes to convery.

See a separate discussion of this point .

Graphic Overview

A Graphic overview is useful and could be built automatically. Should it be made by the author, server, browser or an independent daemon?

Can one provide an overview with less granularity than the basic web by grouping nodes in some way? The user could select from link types used to imply the tree structure. (JFG)

I think this depends on how long it will take. It might be interesting to experiment with daemons which will independently make and update maps of the web. This is not essential for a first pilot model.

History mechanism

This allows users to retrace their steps. Typical functions provided can be interpreted in a hypertext web as follows:
Home
Go to initial node
Back
Go to the node visited before this one in chronological order. Modify the history to remove the current node.
Next
When the current node is one of several nodes linked to the ªbackº node, go to the next of those nodes. Leave the ªBackº node unchanged. Modify the history to remove the current node and replace it with the "next" (new current) node.
Previous
When the current node is one of several nodes linked to the ªbackº node, go to the preceding one of those nodes.
In many hypertext systems, a tree structure is forcibly imposed on the data, and these functions are interpreted only with respect to the links in the tree. However, the reader as he browses defines a tree, and it may be more relevant to him to use that tree as a basis for these functions. I would therefore suggest that an explicit tree structure not be enforced.

(If a default tree is needed by the system for some reason, then we can always use the creation order: when a node is created it is always created with a link to an existing node. Such links, whatever their type, may be used to define a tree. If they are deleted, an alternative link must be chosen to become a tree link.)

If authors want to write a tree structure into their documents, then the words "after", "before" and "above" could be used to mean a static structure.

Index

An Index helps new readers of a large database quickly find an obscure node. Keyword schemes I include in the general topic of indexes. The index must, like a graphic overview, be built either by the author, or automatically by one of the server, browser, or a daemon. The index entries may be taken from the titles, a keyword list, or the node content or a combination of these. Note that keywords, if they are specifically created rather than random words, map onto hypertext ªconceptº nodes, or nodes of special type ªkeywordº. It is interesting to establish an identity relationship between keywords in two different databases -- this may lead a searcher from one database into another.

Index schemes are important but indexes or keywords should look like normal hypertext nodes. The particular special operation one can do with a good keyword index system which one can't do with a normal hypertext system is to do a fast search on multiple keywords. This must to be provided as an extension to the hypertext navigation scheme. However, it is in fact analogous to a trace starting with more than one node, which is a valid hypertext tracing operation. The difference is that the tracing would normally be done by a browser, but the indexed search done by the server.

When many nodes in a web represent different indexes, then a query search can chain between them (See " Web of indexes ").

See also: HyperText and Information Retrieval

Node Names

These allow faster access if one knows the name. They allow people to give references to hypertext nodes in other documents, over the telephone, etc. This is very useful. However, in Notecards, where the naming of nodes was enforced, it was found that thinking up names for nodes was a bore for users. KMS thought that being able to jump to a named node was important. The node name allows a command line interface to be used to add new nodes.

I think that naming a node should be optional: perhaps by default the system could provide a number which can be used instead of a name.The system should certainly support the naming of nodes, and access by name.

Menu of links

Regular linkwise navigation may be done with ªhotspotsº (highlighted anchors) or may be done with a menu. It may be useful to have a menu of all the links from a given node as an alternative way of navigating. Enquire, for example, offers a menu of references as the only way of navigating.