Visual Analytics - Mapping Scientific Frontiers: The Quest for Knowledge Visualization - page 318

Fig. 9.2 CiteSpace labels clusters with title terms of articles that cite corresponding clusters

CiteSpace supports a series of functions that transform the bibliographic data
into interactive visualizations of networks. Users can choose a window of analysis.
CiteSpace divides the entire window of analysis into a sequence of consecutive time
intervals, called time slices. Citation behaviors observed within each time slice are
used to construct a network model. Networks over adjacent time slices are merged
to form a network over a longer period of time.
Synthesized networks can be divided into clusters of co-cited references. Each
cluster contains a set of references. The formation of a cluster is resulted from the
citation behaviors of a group of scientists who are concerned with the same set of
research problems. A cluster therefore can be seen as the footprint of an invisible
college. As the invisible college changes its research focus, its footprints will move
on the landscape of scientific knowledge. The cluster will evolve accordingly. For
example, it may continue to grow in size, branch out to several smaller clusters, or
join other clusters. It may even be phased out as the invisible college drift away from
an old line of research altogether.
CiteSpace provides three algorithms to label a cluster, namely, the traditional
tf by idf, log-likelihood ratio, and mutual information (Fig. 9.2 ). Label terms are
selected from titles, keywords, or abstracts of articles that specifically cite members
of the cluster. If the members of a cluster represent the footprints of an invisible
college or a paradigm, the labels reflect what the invisible college and the paradigm