Tracing Competing Paradigms - Mapping Scientific Frontiers: The Quest for Knowledge Visualization - page 202

of the map's general spatial framework and add the landmarks into the image
subsequently ( 1994 ). Once a mental image is in place, it becomes a powerful tool
for retrieving information. The mental image integrates information about individual
landmarks in a single relatively intact piece, which allow rapid and easy access to
the landmarks embedded. In addition, the greater the integration of structural and
feature information in the image, the more intact the image is. The more intact the
image, the more easily landmark information can be located and help retrieval of
further details.
These findings about thematic maps provide useful design guidelines for in-
formation visualization. In a previous study of visualizing a knowledge domain's
intellectual structure (Chen and Paul 2001 ), we developed a four-step procedure
to construct a landscape of a knowledge domain based on citation and co-citation
data. Our method extracts structural information from a variety of association
measures, such as co-citation, co-word, or co-descriptor. The structural information
is represented as a Pathfinder network, which essentially consists of shortest paths
connecting the network components. The feature information in our visualization
mainly corresponds to citation impact and specialty memberships. Citation impact
of an article is depicted by the height of its citation bar. The color of each year's
citation bar indicates the recentness of citations. Identifying a landmark in such
knowledge landscape becomes a simple task: a tall citation bar with a large amount
of segments in bright color is likely to be a landmark article in the given knowledge
domain. In our approach, the membership of specialty, sometimes also known as a
sub-domain, or a theme, is colored according to the results of factor analysis.
In the following two case studies, we intend to highlight structural and feature
information associated with debates between competing paradigms. We also want
to highlight the movement of a paradigm in terms of the movement of landmark
articles in the global structure. We focus on matching structural and feature
information to what we know about scientific debates involved. A comprehensive
validation with domain experts is a separate topic in its own right.
Kuhn's notion of scientific paradigm indeed provides a framework for us to
match visual-spatial patterns to the movement of an underlying paradigm. If there
exists a predominant paradigm within a scientific discipline, citation patterns should
reflect this phenomenon, allowing the usual delay in publication cycles. A predomi-
nant paradigm should acquire the most citations at least over certain period of time.
Citation peaks are likely to become visible in a landscape view. Two competing
paradigms would show as twin peaks locking in a landscape. Furthermore, such
clusters should be located towards the center of the domain structure. During a
period of normal science, the overall landscape would demonstrate continuous
increases in citations of such clusters. However, if the particular scientific discipline
is in crisis, one or more clusters outside the predominant one will rapidly appear
at the horizon of the virtual landscape. The phenomenon of a paradigm shift takes
place at the moment when the citations of the new clusters of articles take over
that of the original clusters of articles: the peak of the old paradigm drops, while
the valley of a new paradigm rises. Figure 6.2 illustrates the relationship between