A new technique for building maps of large scientific domains based on the co-citation of classes and categories

Abstract

Our objective is the generation of schematic visualizations as interfaces for scientific domain analysis. We propose a new technique that uses thematic classification (classes and categories) as entities of co-citation and units of measure, and demonstrate the viability of this methodology through the representation and analysis of a domain of great dimensions. The main features of the maps obtained are discussed, and proposals are made for future improvements and applications.

Full text

Page 1

Jointly published by Akadémiai Kiadó, Budapest
and Kluwer Academic Publishers, Dordrecht
Scientometrics,
Vol. 61, No. 1 (2004) 129–145
Received April 23, 2004
Address for correspondence:
FÉLIX MOYA-ANEGÓN
University of Granada, Library and Information Science Faculty
Campus Cartuja, 18071 Granada, Spain
E-mail: felix@ugr.es
0138–9130/2004/US $ 20.00
Copyright © 2004 Akadémiai Kiadó, Budapest
All rights reserved
A new technique for building maps of large scientific
domains based on the cocitation of classes and categories
FÉLIX MOYA-ANEGÓN, BENJAMÍN VARGAS-QUESADA, VICTOR HERRERO-SOLANA,
ZAIDA CHINCHILLA-RODRÍGUEZ, ELENA CORERA-ÁLVAREZ,
FRANCISCO J. MUNOZ-FERNÁNDEZ
University of Granada, Library and Information Science Faculty, Granada (Spain)
Our objective is the generation of schematic visualizations as interfaces for scientific domain
analysis. We propose a new technique that uses thematic classification (classes and categories) as
entities of cocitation and units of measure, and demonstrate the viability of this methodology
through the representation and analysis of a domain of great dimensions. The main features of the
maps obtained are discussed, and proposals are made for future improvements and applications.
Introduction
Scientific information is spread out over disciplines which, to the outside observer,
may seem to have little in common. For this reason, when traditional methods are used
to study a domain pertaining to one specific field of knowledge, one is sometimes left
with a sensation of not grasping the domain as a whole. It is like trying to complete a
puzzle and not knowing where to put the piece held in the hand, not seeing which
puzzle pieces it fits in with.
The representation of scientific information in ways easier for the human mind to
embrace is nothing new. To make visible to the mind that which is not visible to the
eye, or to create a mental image of something that is not obvious (e.g. an abstraction)1,
are two definitions of the word “visualization” that point to the intrinsic need to
represent information in a non-traditional manner. To paraphrase Costa2, visualizing is
neither the implicit result of the act of seeing, nor a spontaneous product of the
individual receiving visible input. To visualize is a task of the communicative process,
through which abstract data and complex phenomena of reality are transformed into
visible messages. This enables individuals to apprehend with their own eyes certain data
and phenomena that cannot be directly retrieved from a hidden body of knowledge.

Page 2

F. MOYA-ANEGÓN et al.: Maps of large scientific domains
130 Scientometrics 61 (2004)
The realm of graphic languages for visualizing these “invisible effects” constitutes a
new science of visual communication – schematics – which Costa has defined as the
“third language” after the image and the sign.
The present study proposes a new technique for schematic visualization applied to
the analysis of large scientific domains. The scientific domain is understood in the terms
put forth by Hjørland and Albrechten3, as the reflection of interactions between authors,
and their role in science, through citation. The methodology applied allows the analysis
of data and can serve as an intermediate step in information retrieval.
We shall begin with an overview of different contributions to, or attempts at, the
representation of large scientific domains, to see how they approached the visualization
of small domains while available tools and techniques matured. The elements and units
of measure that could be used for the graphic representation of vast domains are then
discussed, as is the methodology developed for the generation of maps that make
browsing easier. Finally, our results and conclusions are presented, and the particular
features of several examples of maps are mentioned.
A brief bibliographic review of contributions to the representation of
large scientific domains
The first author to articulate this need was Doyle, in 1961.4 He underlined the
importance of computers in producing maps similar to those that the brain generates,
and showed how they can be projected in multidimensional spaces. He also offered his
opinions as to how to construct such maps. To get “the big picture” of a large scientific
domain has allured many researchers since. Garfield,5 in an article published in
American Documentation, showed vivid interest in the construction of historical maps
by means of citation. This interest, taken up by Sher in 1964, led to the creation of a
historical map showing the development of DNA from Mendel to Nirenberg. That same
year Garfield, Sher and Torpie6 generated manually distinct historic-topological science
maps, on the basis of citation of DNA research, using bibliographic coupling as a
variable. DeSolla Price7 showed that the patterns of citations by the authors of scientific
articles would define research fronts, and could also be used to sketch a topology
reflecting the intellectual structure of a scientific domain. But the giant leap forward in
the possibilities for building domain maps or graphs came, in our opinion, from Small8
and Marshakova,9 each of whom proposed the use of document cocitation as a variable
of study in the analysis of citations of scientific production. Science maps showing all
the special fields of the natural sciences, elaborated by Small and Griffith10 or Griffith,
Small, Stonehill and Dey,11 using the Science Citation Index (SCI) as their source of
information and cocitation as a variable of relation, stand as a landmark in the
development of the representation of scientific domains.

Page 3

F. MOYA-ANEGÓN et al.: Maps of large scientific domains
Scientometrics 61 (2004)
131
The most important point about the methodology used by the above authors is that
the groups of documents with common intellectual interests were identified. This was
proof that science is a network of interconnected special fields that can be contemplated
through the quantitative analysis of written production. In 1975, along these same lines,
Aaronson12X-rayed the biomedical publications from the years 1972 and 1973 to
observe their evolution over this time and highlight, on the 1973 map, what the author
called a supercluster, showing the convergence of special fields. The maps by Aaronson
are another turning point in the history of domain representation; not only do they show
its evolution and how the different disciplines mutually interact, but they also supply
information: the documents conform a map of clusters. Each cluster is characterized by
a name, a precise number of documents comprising it, and their degree of interrelation.
This is represented by connecting lines that indicate the value of cocitation. The most
important clusters stand out very clearly, as they have a greater number of relationships
with the rest.
Shortly after the appearance of the Aaronson maps, Garfield13 reported that the
Institute for Science Information (ISI) was working on the elaboration of an Atlas of
Science. This project took six years to materialize. It was finally in 198114 when the first
two volumes of the Atlas appeared, on biochemistry and molecular biology. The
techniques used for the generation of the Atlas maps are quite similar to those used by
Small and Griffith in 1974 (that is, based on the cocitation of documents of a specific
discipline), yet a new spatial positioning of the clusters is achieved by applying
Multidimensional Scaling (MDS). Over time, new volumes of the Atlas of Science have
presented biotechnology and molecular genetics,15 and biochemistry, immunology, and
plant and animal biology,16 then drew to a close, to date at least, after the volume on
Pharmacology.17 Meanwhile, Small has continued to work on the design of maps of
scientific domains, refining the techniques used in his early maps. He can be considered
the ISI´s top specialist in the research and development of science maps.18-25
After the 90’s, new methods of information retrieval and new techniques for the
analysis, visualization and spatial positioning of information (well reviewed by Börner,
Chen and Boyack26), studies based on techniques for visualizing the structure of small
scientific domains begin to proliferate. So, for instance, Braam, Moed and van Raan27,28
propose the combined use of cocitation with co-word analysis for the generation of
science maps, emphasizing their structure and dynamic aspects. Lin, Soergel and
Marchionini29 develop a Self-Organizing Map (SOM) that represents the semantic
relationships among documents and can be used as a bibliographic interface for the
retrieval of online information. Hjørland and Albrechtsen3 put forth a new model based
on the way of thinking or speaking of the society, in turn determined by the social,
economic, or work setting, and which they call domain analysis. White and McCain,30
on the basis of this domain analysis, propose graphic representation or visualization as a
model for information retrieval and analysis. To corroborate their theory, in 1988 they

Page 4

F. MOYA-ANEGÓN et al.: Maps of large scientific domains
132 Scientometrics 61 (2004)
analyzed information science31 in terms of the authors of the 12 journals that, according
to their criteria, were most important between 1972 and 1995. Garfield32states that the
new techniques of visualization make possible the generation of global science maps
which, by zooming over or representing different time periods, allow us to identify the
emerging research fronts, revealing the interests of researchers now at work and
allowing us to associate author names to each front.
White, Lin and McCain33 compare the relatively traditional mode of visualizing
scientific domains using MDS with the Self-organizing Maps (SOM), to conclude that
the latter make it slightly easier to integrate and retrieve bibliographic information.
Ding, Chowdhury, Foo and Qiang34 use bibliometric techniques to break down an area
of knowledge into its main elements, and represent the areas and subareas graphically.
White35 presents networks revolving around on a subject, and proposes that maps be
made with an author name supplied by a non-expert user. This involves a lesser
cognitive load (for the user) the development of interfaces for inexpert users facilitate
the retrieval of information from the bibliographic information, and the possibility that
these interfaces can be generated in a dynamic way. Noyons, Moed and Luwel,36Buter
and Noyons,37 and Noyons38 point to the great potential of science maps as an interface
for information retrieval, and suggest some ways to aid the user in exploring and
comprehending what he sees. Chen and Paul39 describe a method that broadens and
transforms traditional coauthor analysis into structural patterns of scientific literature
that can be represented in 3D maps. For Chen, Paul and O’Keefe,40 the essence of
knowledge, whether in a geographical, thematic, or intellectual context, is the key to
visualization. The proliferation of techniques for the visualization of information allows
the core of knowledge to be represented. This is a key role in the process of
modelization and representation of the structure or intellectual map of a given domain,
be it geographical, thematic or intellectual.
Again, Ding, Chowdhury and Foo41 make a map of the intellectual structure of the
field of information retrieval, this time over a ten-year period (1987–1997), showing
models, patterns and trends of the field as well as different measurements of the degrees
of association among the most relevant terms of the document produced under the
heading “information retrieval.” Ingwersen and Larsen42 use MDS to create a map of
the production of 17 European countries in nine areas of the social sciences. Guerrero
Bote, Moya Anegón and Herrero Solana43classify database documents automatically
using a SOM, and describe how it can be used for browsing and retrieving information
from the database. White, Buzydlowski and Lin,44-46on the basis of the Pathfinder
networks (PFNET) used by Chen,47 and White’s CAMEOs,48 devise a dynamic system
of visualization called Authorlink (using author cocitation) that allows browsing and
information retrieval in real time from a database with records of the Arts and
Humanities Citation Index (A&HCI). Small49 theorizes about the design of a web tool
able to detect and monitor, in real time, the changes on research fronts resulting from its

Page 5

F. MOYA-ANEGÓN et al.: Maps of large scientific domains
Scientometrics 61 (2004)
133
interactions. Chen and Kuljis,50 using citation and cocitation, study the appearance and
evolution of new research fronts in the field of physics. Morris, Yen, Wu and Asnake51
work on visualization, detection and identification of changes on research fronts over
time. Boyack and Börner,52 in an evaluative study, generate maps of scientific
publications with government grants to reveal connections between funding and the
number of citations received.
What we can clearly derive from this general overview is that domain maps or
visualizations are primarily used, thus far, to reveal relationships among documents, to
detect the most important authors within a given discipline, or to analyze the structure
of an area of knowledge and its evolution. The methodology may involve clustering,
MDS, factor analysis, or social networks based on models of graphs, or some
combination thereof.
At present, a great deal of research shares the common objective of producing an
initial map of a domain (a blueprint of sorts) that is general and informative enough to
orient the non-expert user, while also capable of panning over or zooming in on levels
of a discipline using multivariate techniques or network analysis.
Our proposal
In our opinion, cocitation is to date the best tool for obtaining relational information
on documents with which to schematically represent the image of a domain. Thus,
depending on the type of variable analyzed (words, documents, authors, journals), it is
possible to offer different snapshots of a domain to reflect the relationships existing
among the component elements. The use of one variable or another in cocitation will
depend largely on the size of the domain to be represented, as well as on the possibility
or interest in obtaining dynamic visualizations – that is, either online or offline maps.30
But there may also be persistent physical limitations, such as the representation of
information in a low-resolution reduced and static space, which has drawn the attention
of Tufte53,54 for over a decade.
Our proposal entails the graphic and schematic representation of vast thematic,
geographical or institutional domains based on cocitation. It is obviously necessary to
perform some type of clustering, if we hope to corral the intellectual structure of a large
domain on a computer screen, making it quickly intelligible for the human eye and
mind. We propose the cocitation of classes and categories based on the following
scheme:
It is generally accepted by the research community that the frequency with which
any two documents are cocited reflects their degree of affinity in the opinion of the
citing authors.

Page 6

F. MOYA-ANEGÓN et al.: Maps of large scientific domains
134 Scientometrics 61 (2004)
In Figure 1, documents A and B are cocited in document X. The intensity with which
A and B are mutually related depends on the number of times that they are cocited.
Figure 1. Cocitation scheme
Just as the reseach community accepts the fact that documents A and B are an entity of
cocitation valid for the representation of the structure of a domain, making evident their
semantic and intellectual relationships, it also accepts the use of authors as the entity of
cocitation and as the unit of measure for representing this domain. The same happens in
the case of journals.
The ISI, in the Journal Citation Report (JCR),55 places every journal in at least one
subject category. For example, the journal Scientometrics is situated in the category
Computer Science, Interdisciplinary Applications. By analogy, the cocitation of
categories assigned by the ISI-JCR could be used as the entity of measure following the
order established in Figure 1, that is: documents, authors, journals, and ISI-JCR
subjects. To go back to our example, if document A was published in Scientometrics
and therefore has been assigned to the above category, while document B appears in
Pattern Recognition Letters, assigned to the ISI-JCR categories of Computer Science,
Artificial Intelligence, the interrelationship of all these categories becomes evident. The
intensity of the connections will depend on the number of times that documents
published in journals of these subject categories are cocited.
There are different classifications that group ISI-JCR subject categories in superior
conglomerates – heretofore referred to as classes. Because we attempt to represent the
structure of Spain’s scientific production, we have adopted the classification of the
Agencia Nacional de Evaluación y Prospectiva (ANEP),56 which is a taxonomy carried
out by experts in the evaluation of science for the technical and scientific assessment of

Page 7

F. MOYA-ANEGÓN et al.: Maps of large scientific domains
Scientometrics 61 (2004)
135
the action plans of the National Plan of Spanish Scientific Research, adapting the ISI-
JCR categories as a distinctive classifying element of scientific output in Spain, to
twenty-five classes. As is the case in the ISI-JCR, one same category can belong to
different subject areas. Accordingly, the classes may perfectly well be used as the entity
of cocitation and as a unit of greater measure, valid for representing the intellectual
structure of a domain, as we will show later on.
In short, we propose the use of the cocitation of classes and categories as entities of
cocitation and units of measurement for the generation of great schematic visualizations
that act as graphic interfaces for domain analysis.
Methodology
For strictly academic purposes we downloaded from the Web of Science57–
specifically from: the Science Citation Index-Expanded (SCI-E), the Social Science
Citation Index (SSCI) and the Arts and Humanities Citation Index (A&HCI) – the
records with at least one Spanish address in the field “address” from the year 2000, and
put them into an ad hoc database for consultation. The database held 172,562 author
names, who published a total of 26,062 documents (articles, biographical items, book
reviews, corrections, editorial materials, letters, meeting abstracts, news items and
reviews) in 3,838 different ISI journals. When these were broken down into the 243
categories established by the ISI-JCR for the year 2000, 222 categories were covered.
(Spain did not produce any scientific communications pertaining to the remaining 21
categories within the year 2000.) These 222 categories were grouped, based on the
ANEP classification, in 25 classes, again taking into account that one single category
may belong to different classes.
Because we try to show the relationships existing among diverse disciplines in the
natural sciences, social sciences, arts and humanities, we must first solve the problem of
uneven level of citation, as suggested by Small and Garfield.19 For this reason, when
carrying out the cocitation queries corresponding to the classes or categories, we
normalize this measurement of association by dividing the cocitation by the square root
of the product of the frequency of the cites of the cocited documents:
cj ci
Ccij
Cc
·
=
, (Measurement of normalized cocitation58)
where:
– Cc is cocitation
– c is citation.

Page 8

F. MOYA-ANEGÓN et al.: Maps of large scientific domains
136 Scientometrics 61 (2004)
Kamada & Kawai’s algorithm59 was used to automatically produce representations
on a plane, starting from a circular position of the nodes. It generates social networks
with aesthetic criteria such as the maximum use of available space, the minimum
number of crossed links, the forced separation of nodes, building balanced maps, etc.
The result is a tree structure with the following characteristics: a map representing
the Spanish scientific structure as a whole dividing it in 25 big classes. 25 maps, one for
every ANEP class, each of which containing the ISI-JCR categories that ANEP
evaluation experts have considered appropriate. And finally, 222 maps of ISI-JCR
categories, one for every ISI-JCR category with its nearest neighbours.
Class cocitation map (first level)
To obtain the basic data needed for a graphic representation of the whole domain of
Spain, we carry out a cocitation query of classes under the ANEP classification, as
described above.
The result is a symmetrical class cocitation matrix of 25 × 25. Of course, the degree
of intellectual connections shown by the cocitation matrix among certain classes is very
high, making it difficult in some zones to clearly visualize the structure of the domain.
In following the advice of Small,25 therefore, we believe it better to eliminate some
connections, as “the loss of information of the structure implies a gain in simplicity,
justifying the sacrifice in some cases.” We therefore prune the relations between classes
using the Minimum Spanning Tree (MST): when relationships among classes are under
a threshold value, they are successively deleted until only one is left, totally
disconnected from the rest. Then the threshold value is re-established, leaving no class
disconnected. Other available pruning tools include PFNET; but as White60 points out,
this algorithm prunes all the paths except those having the highest degree of meaningful
cocitation, leaving a very reduced number of coincidences in the matrix, where
1
−= nq
and
∞=
r
. Since our aim is to show the highest number of possible links to
represent the structure of a domain as well as its semantic and intellectual relationships,
we opted to use MST, because it allows us to reduce the minimum number of links in
order to facilitate the domain visualization and analysis without loose information.
Similar results might be obtained using PFNET with a lower q, such as
example, but this will be addressed in the future.
We assign each class a different color to distinguish it from the rest, and a size
proportional to the totality of Spanish scientific output in the year 2000, as well as an
ANEP label. This information, together with the cocitation matrix, is processed by the
Kamada & Kawai algorithm to produce a social network where each class is
represented by a node connected with other nodes by undirected links. The relationships
among them and their intensity is seen in the thickness of the links. In just a few cases
in which the tag of a node is partly superimposed on another, we manually modified the
2
=
q
for

Page 9

F. MOYA-ANEGÓN et al.: Maps of large scientific domains
Scientometrics 61 (2004)
137
location assigned by the algorithm a bit. The definitive network is exported to an
Scalable Vector Graphic (SVG) format,61,62 which allows us to zoom in on or shift the
graphics in any direction on the screen.
Category cocitation map (secondary level)
The process involved is very similar to that used for classes, the only difference
residing in details for improving the visualization of the intellectual relationships. As
we commented earlier, each one of the classes is made up of a specific number of ISI
categories, assigned according to the criteria of the ANEP. For each ANEP class, we
consult the cocitation of ISI categories, normalized as explained earlier, to obtain a
symmetrical cocitation matrix of n × n categories, based on the number of each in each
class. After pruning by MST we assign a color to each category, which is the color of
the class it belongs to. We adjust slightly in each level: the category with the greatest
scientific output is the biggest one, and the rest are made proportional to this, reflecting
their relative magnitude in the context of total publication. We observe that there is little
difference among the categories of a single class, and those with a small yield, such as
four works, are perfectly visible.
The Kamada & Kawai algorithm is supplied with the name of the categories that
make up each class, their size, color, and the corresponding cocitation matrix, which is
what establishes the relationships among categories. In this case, we indicate to the
algorithm that these relationships are to be directed. Though it is a symmetric matrix
(and therefore when relationships exist among categories, the categories will likewise
be symmetrical), taking this small license helps us achieve a very clear and well
structured visualization of the domain. The thickness of the links again indicates the
intensity of their interrelationships.
The social network that we obtain, modified only in a few cases to avoid the overlap
of tags on nodes or categories, is exported to an SVG format.
Map of neighbors (third level)
Here we do not take into account the ANEP classification, but use the ISI categories
alone. Because we want to discover the documents hidden behind each category and
each link, we create an egocentered star-shaped network of all the categories with
scientific output from Spain in 2000.
We depart from a cocitation matrix of 222 × 222 categories. From there, we build a
list of neighbors based on the specific subject area under consideration. We use the
Kamada & Kawai algorithm to process the list of neighbors, the name of each category,
color and size. As it would be impossible to clearly show the 222 categories, we prune,
but this time not with MST. Rather, we take as the threshold the figure obtained from

Page 10

F. MOYA-ANEGÓN et al.: Maps of large scientific domains
138 Scientometrics 61 (2004)
calculating the average plus the standard deviation of those categories which, within the
complete cocitation matrix, belong to the vector of the category that we are going to
represent, and which have a value greater than zero. In other words, we use a pruning
mechanism with a variable threshold, established by the idiosyncrasy of the links of
each central category itself, with respect to the rest. We eliminate the links and vertices
that are connected to the central node or category, with a threshold value under the
average plus standard deviation. To represent this as a network we again use the
Kamada & Kawai algorithm, but now specifying that the value of the distance between
vertices is a similarity function of cocitation. Hence, the thickness of the lines is always
the same, while their length varies. It is thus obvious which categories are closest to the
central one, and share a greater topic affinity. In some instances, when links or node
tags overlap, they are manually shifted a bit. The map obtained is exported to the SVG
format.
Results
We believe that this type of representation allows any user to grasp, clearly and
quickly, the structure of a domain by observing its nodes and links, and their thickness
or distance.
Map of the first level or class cocitation
Figure 2. Class cocitation map

Page 11

F. MOYA-ANEGÓN et al.: Maps of large scientific domains
Scientometrics 61 (2004)
139
Each class is represented by a node in the network with its name. At first glance,
Spain’s scientific output for the year 2000 is widely distributed, over the length and
width of the graphic area. We note that there is a large “gap” in the center, indicating
there is no subject area associated directly with all the others. This is also what happens
with MDS representations. Even Multidisciplinary, the class with the greatest number
of relationships, is found at a roughly left-center position, closer to the area of Sciences.
From a general standpoint, we can clearly distinguish the three ISI databases. In the
leftward area SCI-E is vaguely contoured, wheras in the upper right the SSCI database
is reflected, and pendent from it, the contents of what would be the A&HCI. Within the
SCI-E zone, there are three prominent blocks: one we could call Life Sciences,
including Livestock and Fishing, Food Sciences and Technology, Medicine, Physiology
and Pharmacology, Psychology and Educational Sciences, Molecular & Cellular
Biology & Genetics, Plant & Animal Biology & Ecology, and Agriculture. Another
block would be that of Physics, Chemistry and Earth and Space Sciences, containing
Chemistry, Geosciences, Chemical Technology, Physics and Space Sciences and
Materials Science & Technology. Finally, the group of Engineering and Computer
Sciences would contain Civil Engineering & Architecture, Computer Sciences &
Technology, Electric, Electronic & Automated Engineering, Mechanical, Naval and
Aeronautical Engineering, Mathematics and Electronic & Telecommunications
Technology. We have deliberately left the Multidisciplinary Sciences out of this
classification because its contents could belong to all three blocks. In the upper right
zone of Social Sciences and Art and the Humanities we quickly infer the SSCI database
represented by Social Sciences, Economy and Law, as well as the A&HCI database,
with History & Arts and Philology & Philosophy. It is noteworthy that Psychology and
Educational Sciences, Mathematics and Social Sciences act as a bridge for the network
as a whole, connecting the three major component groups, which coincide with the
three ISI databases.
The nodes with a greater number of links occupy more or less central positions
easily related with the rest (for example Multidisciplinary Sciences, Physics & Space
Sciences or Chemistry), whereas those with fewer links are situated in the periphery,
(among others, Philology & Philosophy, Economy, Law and even Medicine). At a
glance we note that the link between Social Sciences and Economy is thicker than
Social Sciences’ links with History & Arts, Law, Psychology & Educational Sciences
or Mathematics. Similarly, in the lower part of the Graph, this happens with Electronic
& Telecommunications Technology, Electric, Electronic & Automated Engineering,
Computer Sciences & Technology, Civil Engineering & Architecture and Mechanical,
Naval and Aeronautic Engineering. The thicker links serve as evidence that the use of
common sources is greater.

Page 12

F. MOYA-ANEGÓN et al.: Maps of large scientific domains
140 Scientometrics 61 (2004)
Maps of the second level or cocitation of categories
Each one of the nodes in the class cocitation maps gives rise to the generation of a
new map of ISI category cocitation, which gives us a total of 25 maps, as explained in
Methodology. For reasons of research affinity, we shall use as an example the map of
Social Sciences, which includes the category Library and Information Sciences.
Figure 3a. Category cocitation map: Social Sciences
Figure 3b. Category cocitation map: Social Sciences

Page 13

F. MOYA-ANEGÓN et al.: Maps of large scientific domains
Scientometrics 61 (2004)
141
Every node in the network represents an ISI-JCR category. The numeral alongside is
the number of published works in that field produced by Spanish research institutions
within the year 2000. In the case of the Social Sciences, the most central nodes are, in
this order: Sociology, Planning & Development, Social Sciences-Mathematical
Methods, Social Sciences-Interdisciplinary and Management. The rest, including
Library & Information Sciences, are in more or less peripheral positions.
As in class cocitation maps, the intensity of the links is shown by their thickness. In
the case of the maps of category cocitation alone, however, we have included a small
application that allows us to hide the links in order to enhance visibility. The links
become visible when the cursor is positioned over one of the nodes or tags, as seen in
Figures 3a and 3b.
Of the six links that Library & Information Sciences shares with other nodes, two
are very weak – those with Operation Research & Management Science and Social
Sciences-Interdisciplinary – while the other four are somewhat stronger. Sociology has
fifteen different relationships, with two particularly strong ones: Social Sciences-
Interdisciplinary and Industrial Relations & Labor.
Neighbor maps
Each node from the category cocitation maps becomes the starting point of a new
map, giving rise to a total of 222 graphs we shall call Neighbor Maps. Tracking down
Library & Information Sciences once again, we show its neighbors in Figure 4.

Figure 4. Neighbor map. Social Sciences

Page 14

F. MOYA-ANEGÓN et al.: Maps of large scientific domains
142 Scientometrics 61 (2004)
The characteristic feature of this map type is that it depicts an egocentered network,
where the node studied is always situated in the center, and the rest orbit around it.
Although the representation is balanced and tends to occupy all the available space, the
intensity of the relationships is reflected here by the distance between nodes. Thus, the
most closely related categories are, respectively, Computer Sciences & Information
Systems, Communication, History & Philosophy of Science, Management, Computer
Sciences & Interdisciplinary Applications, Planning & Development, Business and
Social Sciences-Interdisciplinary. The central node or category “attracts” those nodes
with which it maintains a closer relationship (in terms of common sources) regardless of
the ANEP class it belongs to. This makes it very easy to perceive category groupings
and their intensity of correlation without referring to any external classification system.
Conclusions
We have shown that the cocitation of classes or categories can be used to build maps
of large geographic domains. This methodology could be applied to comparative studies
of similar domains with a geographic reference, or even with a thematic or institutional
point of reference. Moreover, the evolution of research fronts within a single domain
can be inferred through sequences of representations over time.
In our view, domain analysis is best organized as a hierarchy that would begin with
maps of classes, followed by category maps, and then by neighbor maps. The logical
continuation of this proposal would be for the latter to be followed by maps of journal
cocitation and author cocitation, in that order. Our current work takes us in this
direction.
While it is true that we present off-line maps here, we are also testing a set of tools
that would allow the automatic generation of schematic representations for a domain the
size of Spain. However, we must remember that the re-generation of these maps is
required as new information is added to the database (to date, annually). At some future
date we may even have the means for dynamic updatings of such maps. For the time
being, however, a “demo” or thematic prototype of Spain is available online, as is as a
prototype of the institutions of Andalucía, for any interested browsers at
www.atlasofscience.net.

Page 15

F. MOYA-ANEGÓN et al.: Maps of large scientific domains
Scientometrics 61 (2004)
143
References
1. Framework of a Visualization System [Web page], Available at:
http://www.siggraph.org/education/materials/HyperVis/abs_con1/main.htm [Review at: 08/08/2003], (1999).
2. COSTA, J., La esquemática: visualizar la información, Barcelona, Paidós, 1998.
3. HJØRLAND, B., ALBRECHTSEN, H., Toward a new horizon in information science: domain analysis,
Journal of the American Society for Information Science (JASIS), 46 (1995) 400–425.
4. DOYLE, L. B., Semantic roadmaps for literature searchers, Journal of the Association for Computing
Machinery, 8 (1961) 553–578.
5. GARFIELD, E., Citation indexes in sociological and historical research, American Documentation, 14
(1963) 289–291.
6. GARFIELD, E., SHER, I. H., TORPIE, R. J., The Use of Citation Data in Writing the History of Science,
Philadelphia, Institute for Scientific Information, (1964).
7. PRICE D., J. D., Networks of scientific papers, Science, 149 (1965) 510–515.
8. SMALL, H., Co-citation in the scientific literature: a new measure of the relationship between two
documents, Journal of the American Society for Information Science (JASIS), 24 (1973) 265–269.
9. MARSHAKOVA, I. V., System of document connection based on references, Nauchno-Teknichescaya
Informatisya, Series II (6) (1973) 3–8.
10. SMALL, H., GRIFFITH, B. C., The structure of scientific literature, I: Identifying and graphing
specialyties, Science Studies, 4 (1974) 17–40.
11. GRIFFITH, B. C., SMALL, H., STONEHILL, J. A., DEY, S., The structure of scientific literature, II: Toward a
macro and microstructure for science, Science Studies, 4 (1974) 339–365.
12. AARONSON, S., The footnotes of science, Mosaic, 6 (March-April) (1975) 22–27.
13. GARFIELD, E., ISI’s Atlas of Science may help students in choice of career in science, Current Contents,
29 (July 21) (1975) 5–8.
14. GARFIELD, E., Introducing the ISI Atlas of Science: Biochemistry and molecular biology, 1978-80,
Current Contents, (42) (1981) 5–13.
15. GARFIELD, E., Introducing the ISI Atlas of Science: Biotechnology and molecular genetics, 1981/82 and
bibliographic update for 1983/84, Current Contents, (41) (1984) 3–15.
16. GARFIELD, E., The encyclopedic ISI-Atlas of Science launches 3 new sections: bochemistry,
inmunology, and animal and plant sicences, Current Contents, (7) (1988) 3–8.
17. SEIDEN, L. S., SWANSON, D. R., ISI Atlas of Science: Pharmacology 1987, Vol 1, Library Quarterly, 59
(1989) 72–73.
18. SMALL, H., The relationship of information science to the social sciences: a co-citation analysis,
Information Processing & Management, 17 (1981) 39–50.
19. SMALL, H., GARFIELD, E., The geography of science: disciplinary and national mappings, Journal of
Information Science, 11 (4) (1985) 147–159.
20. SMALL, H., SWEENEY, E., Clustering the science citation index using co-citations. 1. A comparison of
methods, Scientometrics, 7 (1985) 391–409.
21. SMALL, H., SWEENEY, E., GREENLEE, E., Clustering the science citation index using co.citations. 2.
Mapping science, Scientometrics, 8 (1985) 321–340.
22. SMALL, H., Macrolevel changes in the structure of cocitation clusters: 1983-1989, Scientometrics, 26
(1993) 5–20.
23. SMALL, H., A SCI-MAP case-study: building a map of aids research, Scientometrics, 30 (1994)
229–241.
24. SMALL, H., Visualizing science by citation mapping, Journal of the American Society for Information
Science (JASIS), 50 (1999) 799–813.
25. SMALL, H., Charting pathways through science: exploring Garfield’s vision of a unified index to science,
In: B. CRONIN, H. B. ATKINS (Eds), The Web of Knowledge: A Festschrift in Honor of Eugene Garfield.
Medford, N. J. Information Today, 2000, pp. 449–473.
26. BÖRNER, K., CHEN, C., BOYACK, K. W., Visualizing knowledge domains, Annual Review of Information
Science & Technology, 37 (2003) 179–255.