Professional Documents
Culture Documents
Elena I. Gaura
Robert M. Newman
Coventry University
Coventry University
E.Gaura@coventry.ac.uk
R.M.Newman@coventry.ac.uk
ABSTRACT
2.
3.
2. USABILITY OF STRUCTURE
For the purposes of this discussion we are interested in the
usability of hypermedia structure, abstracted away from the more
commonly considered usability issues of presentation or content.
The proposition put forward here is that structure, in terms of the
connectivity or graph of a hypermedia document can affect its
usability. The connection between structure and usability has been
noted for some time, starting with the definition of the
subsequently much used term lost in hyperspace [1].
Surprisingly, however, there appears to have been relatively little
work done so far to find out definitively how structure affects
usability. Within this field, the major work on usability of
structure was by Botafogo et al., [2] who developed metrics for
usability of structure. The metrics developed classified nodes
using their distance (in links traversed) from other nodes. The
measures used are Relative In Centrality (RIC) and Relative Out
Centrality (ROC), which is a normalised measure of the distance,
in terms of links, from other nodes. Mukherjea and Foley [3] have
developed a method to identify landmarks based also on
connectedness with other nodes. Modjeska and Marsh have
investigated the effects of hierarchical structure on hypermedia
usability, finding that generally hierarchical sites offer greater
usability whereas more weakly hierarchical sites offer faster
navigation, but find that landmarks have little effect on usability.
General Terms
Documentation, Design, Theory.
Keywords
Hypermedia, Artificial Intelligence, Artificial Neural Networks.
1. INTRODUCTION
In the design of large, closed hypermedia systems design and
control of structure is one of the major issues. It is known that
different structural attributes impart different usability
characteristics. At the same time, from the designers point of
view, it has been observed that it is a mistake to impose an
arbitrary structure too early or too late in the process. Existing
design techniques for large hypermedia systems generally address
this problem using one, or a combination of, three techniques.
1.
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies
are not made or distributed for profit or commercial advantage and
that copies bear this notice and the full citation on the first page. To
copy otherwise, or republish, to post on servers or to redistribute to
lists, requires prior specific permission and/or a fee.
SIGDOC03, October 1215, 2003, San Francisco, California, USA.
Copyright 2003 ACM 1-58113-696-X/03/0010$5.00.
100
2.
3.
4. ENCODING STRUCTURE
In order to apply ANNs to the recognition of desirable or
undesirable structure a means must be devised for encoding
structure into a form suitable for their input.
Gogan and Buranga [11] describe such a means. Their method of
encoding was simply to count the links in the top, middle and
lower portion of a document. Although this provided a suitable
basis for classification using a self organising map (SOM,
described in more detail below), such an encoding method seems
to be of experimental value only, it is difficult to see how it might
be of practical use.
Bollen and Heylighen [8] note the way in which the World Wide
Web, as a whole, has developed in a very unstructured way, and
that the quality of its structure does not seem to improve. They
propose development of automatic learning algorithms to allow
hypermedia systems to adapt their own structure by learning the
101
counting its inputs and outputs the type of structure that it resides
in. For example, consider the tree structure below.
Net density corrected for the fact that large sites have more
potential links per page, and is the ration between actual number
of links and the theoretical maximum.
Compactness the total distance between all pages compared to
the theoretical maximum.
Linearity the amount that the order of reading is imposed by the
author, expressed by Botafogos stratum metric.
Circumference a graphs longest cycle.
102
links is not so simple. The first page shown is a home page, and
fulfils Botafogos proposals concerning how they be defined.
However, it is also a content page, because it delivers an amount
of general information. Hybrid pages such as this are common in
modern hypermedia, mainly because they maximize the
information imparted to navigation distance ratio. However, when
we look at another page from the same site the position becomes
more complex.
103
7. CONCLUSION
Neural networks have been used widely in information retrieval
and data mining, and are therefore common on the user side of
hypermedia. The authors believe that they also have great
potential for aiding the authoring side, and in a more flexible and
less absolute way than traditional design aids based on the
discrete mathematics of structure and connectivity. This paper has
represented the initial consideration of this potential which forms
the start of a range of investigations in this area. Subsequent work
will explore the use of ANNs for discovery of structure and its
classification within hypermedia and for the implementation of
authoring aides to help designers keep track and control of
structure.
8. REFERENCES
[1] Edwards, D., Hardman, L., Lost in Hyperspace.
Cognitive mapping in a hypertext environment. In
McAleese, R. ed., Hypertext, theory into Practice.,
Intellect, Oxford, 1993
104