|
Items in cart [0] |
Questions? Call 888-624-8373 |
| Copyright © 2009. National Academy of Sciences. All rights reserved. Terms of Use and Privacy Statement |
Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 109
Colloquium
Crossmaps: Visualization of overlapping relationships
in collections of journal papers
Steven A. Morris* and Gary G. Yen
Electrical and Computer Engineering, Oklahoma State University, 202 Engineering South, Stillwater, OK 74078
A crossmapping technique is introduced for visualizing multiple
and overlapping relations among entity types in collections of
journal articles. Groups of entities from two entity types are
crossplotted to show correspondence of relations. For example,
author collaboration groups are plotted on the x axis against
groups of papers (research fronts) on the y axis. At the intersection
of each pair of author group/research front pairs a circular symbol
is plotted whose size is proportional to the number of times that
authors in the group appear as authors in papers in the research
front. Entity groups are found by agglomerative hierarchical clus-
tering using conventional similarity measures. Crossmaps comprise
a simple technique that is particularly suited to showing overlap in
relations among entity groups. Particularly useful crossmaps are:
research fronts against base reference clusters, research fronts
against author collaboration groups, and research fronts against
term co-occurrence clusters. When exploring the knowledge do-
main of a collection of journal papers, it is useful to have several
crossmaps of different entity pairs, complemented by research
front timelines and base reference cluster timelines.
Collections of journal papers related to a scientific field are a
useful source of information when mapping a knowledge
domain (1~. The structure within the knowledge domain is mani-
fested in the collection of papers as groups of related entities, such
as groups of papers that represent subtopics, groups of references
that represent base knowledge, groups of paper authors that
represent collaboration teams, groups of reference authors that
represent experts, groups of journals that represent subtopic librar-
ies, and groups of terms that represent specialized vocabularies
within the knowledge domain. Exploration and visualization of
these groups and the complex relations among them provides
information that can be used to gain a broad and detailed under-
standing of the underlying knowledge domain.
Inherently, entity groups within collections of journal papers
exhibit considerable "core and scatter" in group membership
(2), with each group usually possessing a small core group of
strongly related member entities and a much larger group of
weakly related scatter members. Furthermore, there is consid-
erable overlap in membership of entities in groups. For a
thorough understanding of the structure of a knowledge domain,
it is useful to visualize and understand the extent of overlap
among groups in a collection of journal papers.
This article introduces a simple technique for visualizing the
relations among collections of entity groups. The technique,
which uses a crossmap format to show the magnitude of corre-
spondences between all pairs of groups drawn from two differing
entity types, allows visualization of relations between groups and
additionally permits visualization of overlap in group member-
ship. Using this technique, it is possible to visualize and under-
stand the set of complex relations among the different groups
that are manifested in a knowledge domain. For example, given
a collection of research fronts, i.e., groups of papers reporting on
the same subtopic, for each research front it is possible to identify
the groups of important references, contributing author collab-
www.pnas.org/cgi/cioi/10. 1 073/pnas.03076041 00
oration teams, groups of experts (important reference authors),
and key journals.
This article is organized as follows. The reader is introduced
to a simple entity-relationship model of collections of journal
papers. This is followed by a discussion of important entities used
for mapping knowledge domains and a discussion of co-
occurrence relations that are used to cluster entities into groups.
A detailed discussion of important entity groups appears, in-
cluding an explanation of core and scatter and overlap of group
membership. Then, the use of correspondence metrics between
entity groups is discussed, followed by a detailed discussion of
the proposed crossmapping technique. Finally, an example is
presented, showing crossmaps produced from a collection of
journal papers related to the subject of complex networks.
Entity Model of Journal Paper Collections
Using an entity-relationship model (3), collections of journal
papers may be considered to be a collection of entities of
differing entity types. Examples of entity types for journal paper
collections include papers, paper authors, references, and paper
journals. Borner, Chen, and Boyack (1) describe these entities as
"units of analysis" and list the entity types most commonly used
for mapping knowledge domains and also show applications of
the analysis of each entity type. This article will expand on
analysis of entities for knowledge mapping to include analysis of
relations between different types of entities, thus extending the
understanding of complex knowledge domains.
Within the collection of journal papers, entities are associated
with each other. For example, each paper in the collection is
associated with the authors who wrote it, the references it cites,
the journal in which it was published, and the terms that appear
in it. As presented here, these associations are always between
pairs of entities of differing entity type. Entities of the same
entity type are never directly associated. For example, papers
and references are considered distinct entity types, even though
references often correspond to actual papers. This separation
into two distinct entity types is necessary, because papers and
references represent differing concepts. A paper represents a
research report, whereas a reference represents a symbol of
knowledge (4~. Similar considerations require designation of
paper journals, reference journals, paper authors, and reference
authors as separate entity types. Fig. 1 shows an ontology
diagram of the entities within a collection of journal papers and
the types of associations among those entities. Fig. 2 explains the
symbols used in Fig. 1.
Co-Occurrence Among Entities
Co-occurrence relations among entities of the same entity type
occur when two entities of the same type are associated with an
This paper results from the Arthur M. Sackier Colioquium of the National Academy of
Sciences, "Mapping Knowiec~ge Domains," held May 9-11, 2003, at the Arnold anc] Mabel
Beckman Center of the National Academies of Sciences and Engineering in Irvine, CA.
*To whom corresponclence shouIc] be aciciresseci. E-maii: samorri~tokstate.edu.
2004 by The National Acaciemy of Sciences of the USA
PNAS 1 April 6, 2004 1 vol. 101 1 suppl. 1 1 5291-5296
OCR for page 110
1
INSTITUTION MEMBERSHIP
LISTS
TERMS
RELATED TERMS
(BY COOCCURENCE
IN PAPERS)
THESAURUS ENTRIES
< CONTAINS MULTIPLE MULTIPLE
APPEAR MULTIPLE IN MULTIPLE_
-
ZZP~'S
LAW
C o
BRADFOQD'S m Z
LAW '
io
m
AUTHORS, PAPER
.' L
AUTHOR
COLLABORATION GROUPS
/ \
c ~ ~ ~ LOTKA'5 /QEFEQENCE ° o
m in ~ / POWER [A W m ~
z ~ / (NAQANAN, 1971) ~ Cal
c m / / m
\ / PAPERS /
. ,
COAUTHOR GROUP OEUVRES /
RELATED PAPERS BIB COUPLING
BY COOCCURENCE (RESEARCH
OF TERMS FRONTS)
PAPERS GROUPED BY
PAPER JOURNAL
\
of
_
co
0
Z
_ \
.
NO GROUP
AUTHORS,
REFERENCE
AUTHOR CO-
CITATION
GROUPS
CONTAINS MULTIPLE UNIQUE /
~ >
< APPEARS ONCE IN MULTIPLE
JOURNALS, PAPER
NO GROUPS |
7 0
m
_ \ / REFERENCES
REF ~ GROUPED BY REF AUTHOR l
CO-CITATION
(BASE REFS)
REFS GROUPED BY REF JOURNAL
/\
m of O D
I'm z JOURNALS,
~ \ /REFERENCE
| NO GROUP
Fig. 1. Ontology diagram of the entities within a collection of journal papers, their direct relations to each other, co-occurrence groups, and core and scatter
relations among those entities.
entity of a differing entity type. For example, two authors are
related when they coauthor a paper, two papers are related when
both cite the same reference, or two references are related when
they are cited together in the same paper. Co-occurrence
relations between pairs of entities often imply some meaningful
relation between those entities. For example, coauthorship of
papers implies that pairs of authors are collaborators, common
references between papers implies that pairs of papers deal with
the same research topic, and cocitation of references implies that
two references are symbols of similar base knowledge.
Several of the co-occurrence relations that occur within
collections of journal papers have been named and extensively
studied. These relations are noted in Fig. 1 and include:
Bibliographic coupling: Relation of pairs of papers by com-
mon references, implying a common research topic between
the papers (54.
Cocitation: Relation of pairs of references by their co-occur-
rence in papers, implying that those two references are
symbols of similar base knowledge (6~.
Author cocitation: Relation of pairs of reference authors by
their co-occurrence in papers, implying that the two authors
are symbols of the same base knowledge (7~.
Coauthorship: Relation of pairs of paper authors by coau-
thorship of papers, implying that the two authors are members
of the same collaboration team Gil.
Many other co-occurrence relations are possible, as noted in
Fig. 1. For example, pairs of papers related by common terms
may imply a common research topic, or pairs of journals that
contain papers that cite common reference authors may imply
5292 1 www.pnas.org/cgi/doi/10.1073/pnas.0307604100
that those two journals publish papers that have a common
research topic.
Entity Groups
Using similarity metrics derived from co-occurrence counts
between pairs of entities, and applying clustering techniques,
groups of entities possessing commonalities can be identified in
the collection of journal papers. Examples of commonly studied
groups of entities are noted in Fig. 1 and include (2~:
Research fronts: Groups of papers that share a common
research topic (9~. Derived from co-occurrence of references
in papers, these groups can be considered as representing
Kuhnian puzzles within a scientific field (10, 11~.
Base reference groups: Groups of references that serve as
symbols of similar base knowledge (12~. Derived from co-
occurrence of references in papers, these groups can be
considered as representing Kuhnian exemplars or paradigms.
Reference author groups: Groups of reference authors that
serve as symbols of similar base knowledge (13~. They are
derived from co-occurrence of reference authors in papers.
Similar to base reference groups, these groups can also be
considered as representing Kuhnian exemplars and para-
digms, but on a more abstract scale. Reference author groups
can also be considered as groups of experts (14).
- Collaboration teams: Groups of paper authors that work
together. Derived from coauthorship of papers by paper
authors, these groups can be considered as representing
"invisible colleges" within a field (15, 16~.
· Vocabularies: Groups of keyword terms. Derived from
co-occurrence of terms in papers, these groups can be
Morris and Yen
OCR for page 111
Entity-type to entity-type
relation; how entity at base
of arrow is related to ~ c
entity-type at tip of arrow. \~` An
m ct
m
TERMS
:RMS
.NCE
ERS)
/
CONTAINS MULTIPLE MULTIPLE
<
APPEAR MULTIPLE IN MULTIPLE
77.PF'S
LAW
RELATED PAPERS BIB COUPLING
BY COOCCURENCE (RESEARCH
no TF~M.R FRONTS)
, PAPERS GROUPED BY
PAPER JOURNAL
Observed d istribution B~ADFO~:
| NO GROUP
of entity-type to entity-type I LAW
relation.
... .
a
m
Fig. 2. Key explaining the notation in the ontology diagram of Fig. 1.
considered to represent specialized vocabularies within a
research field (17~.
Core and Scatter and Overlap of Group Membership
Entity groups within collections of papers exhibit core and
scatter. Groups tend to possess a small set of core members that
are strongly related to each other and a large number of scatter
members that are weakly related (2~. Furthermore, weakly
related member entities are ambiguously related to many groups
simultaneously. There is extensive overlap in group membership,
leading to great difficulties when visualizing the knowledge
domain represented by a collection of journal papers.
The core and scatter relations among entities in collections of
journal papers manifest themselves as power-law distributions of
entity frequency. Some of these relations, noted in Fig. 1, have
been extensively researched. Example power-law relations are:
· Lotka's Law for author frequency (184.
· Bradford's Law for paper journal frequency (19~.
· Zipf's Law for frequency of terms (20~.
· Reference power law for frequency of references (21~.
Standard clustering techniques, such as hierarchical agglom-
erative clustering, and standard visualization techniques, such as
multidimensional scaling, do not effectively reveal the overlap of
entity group membership in collections of journal papers. The
crossmapping technique proposed here is designed to reveal this
overlap in a field's knowledge structure by showing overlap in
correspondence among groups taken from differing entity types.
Correspondence Between Groups of Entities from Differing
Entity Types
Define a correspondence metric to measure the relation between
a pair of entity groups drawn from differing entity types. As an
example, a possible correspondence metric between a research
front (a group of papers) and a base reference group is the
percentage of references in the base reference group that is cited by
papers in the research front. Given two collections of groups, each
collection drawn from a different entity type, it is possible to build
Morris and Yen
LOT~A'S /
LAW /
, PAPERS
Entity-type
AUTHOR ~
COLLABORATION GROUPS / | Nt
_ \ /
O ~
COAUTHOR GROUP OEUVRES
CONTAINS MULTIPLE UNIQUE
APPEARS ONCE IN MULTIPLE
<
-\
1
0 g
m A
C ~
Z Z
oh
m
REFS GRO
CO-CITATIO
(BASE REFC
REPS GROl
\ Entity groups formed by
co-occurrence with entity-type
JOURNALS' PAPER at opposite end of relation
a matrix of the correspondences that exist from each group of the
first entity type to each group of the second entity type.
Entity groups overlap in correspondence between groups
from different entity types e.g., a base reference cluster may
have correspondence to several research fronts, or an author
collaboration group may have correspondence to many term
co-occurrence clusters. Knowledge of the correspondence be-
tween groups drawn from different entity types is helpful for
mapping the knowledge domain associated with the collection of
journal papers. Furthermore, visualization of overlapping group-
to-group correspondence helps sort out complex relations
among research topics, base reference groups, and research
teams. In collections of journal papers, we propose several
correspondence relations between groups of different entity
types that are useful for knowledge mapping:
- Relation of research fronts to base reference groups. This
shows what base knowledge supports specific research topics.
· Relation of research fronts to author collaboration groups.
This shows what research teams work on specific research
topics.
· Relation of research fronts to term co-occurrence groups. This
shows what concepts are associated with specific research
topics and can be helpful for labeling research fronts.
· Relation of research fronts to paper journal groups. This
shows the core journals that publish papers pertinent to
specific research topics.
The crossmapping technique presented here is used to visu-
alize and explore relations between groups of entities. The
technique is especially suitable to the visualization of overlap in
such relations, and as such, allows the investigation of a knowl-
edge domain through various manifestations: research fronts,
base reference groups, invisible colleges, technical vocabularies,
and core journals.
Description of Crossmap Visualization
The crossmapping technique presented here visualizes the ma-
trix of correspondence magnitudes between groups from two
PNAS 1 Aprii 6, 2004 1 volt. 101 1 suppl. ~ 1 5293
OCR for page 112
Amaral: 'Classes of
~ small world networks' V .° ~ °o° ooo O=.oo
Jeong: 'The large-scale ~ ~ _
organization of metabolic networks' _to.oO.~3O'~o °coOO s
............ - ^~ P°~°°~°O° If
Albert:'Internet- diameter ~
of the world wide web'
.. . . . . . . . . . . . . . . . . \ . . . . . . . . . . .
Watts:.'Collective... \
dynamics of small \ o -o°~°a,° °oo
world networks' ~
.. ,, . .~. - ~ 'O
. ~ ~.~ ~
.. .... . . . . . ... ... / . .. . .~? ~ ... `IOQ° CI .' °°°q43
Barabasi:'Emergenceof scaling /
in random networks' /
-- -° - - 0 - ~ o O - - O' / ° - - ~ be- (3 Coo O ~ ~ .-° o8 ° °° --
Albert: 'Error and attack tolerance
of complex networks'
1998 1999 2000 2001 2002 2003
DATE
......... - -Q O -8 o<~80O° o ~
.. -....,9 ........
epidemics
eDidemics
. . his n~hN~rk.c
Fig. 3. Timeline of research fronts for complex networks papers. Papers are
shown as circles whose size is proportional to total citations received. Filled
circles are papers that have received eight or more citations in the last 12
months.
different entity types. Assuming, for example, groups of papers
(entity type 1), and groups of references (entity type 2), one
measure of correspondence is the number of references in a
group of references that appears in a group of papers. Given No
groups of entity type 1 and N2 groups of entity type 2, a N~-by-N2
matrix lists all of the correspondences between entity type 1
groups and entity type 2 groups. A crossmap is a visual repre-
sentation of that correspondence matrix. The crossmap method
is similar to MATRIX BROWSER (22), used for visualizing com-
puter networks; DOCCUBE (23), which uses 3D matrix visualiza-
tions to aid query searches of large document collections; and
GRlDL (24), an interactive system for visualizing hierarchically
organized databases and library search results. The crossmap
technique complements timeline visualization (11), allowing a
L
thorough exploration of the static relations and temporal events
in a collection of research fronts.
Construction of Crossmaps
To start, clustering is performed on each entity type, grouping
entities according to some similarity metric. Clusters from the
first entity type are mapped as rows, and clusters of the other
entity type are mapped as columns. Dendrograms are added to
the crossmap to show the structure of clusters being displayed.
For every group at row ~ from entity type 1, and every group at
column j from entity type 2, a circle is placed at row i and column
j whose size is proportional to the magnitude of the correspon-
dence between those two groups. Group labels are placed at row
and column positions to the left and bottom of the map.
Example: A Collection of Complex Networks Papers
A collection of papers about complex networks will illustrate the
use of crossmap techniques. This collection was gathered from
the Institute for Scientific Information Web of Science product
(www.isinet.com/products/citation/wos) by using queries to
gather papers that cite several key references in the field. All
groups in this set were generated by using agglomerative hier-
archical clustering with Ward's method linkage on co-
occurrence metrics normalized by using the cosine formula (25~.
Summary statistics for the entities in this collection are:
· 323 papers in 86 journals (20% of journals contain 76% of all
papers).
· 11,304 citations to 6,167 references (20% of references re-
ceived 54% of all citations).
· 826 authorships of 455 authors (20% of authors accounted for
52% of all authorships).
Using bibliographic coupling (5) as a co-occurrence metric, a
clustering of 10 research fronts was generated. Fig. 3 shows a
timeline of the research fronts. Papers are shown as circles
plotted by publication date in horizontal tracks whose vertical
position corresponds to positions on the clustering dendrogram
shown to the left. Circle size is proportional to the number of
citations received, and circles are darkened for papers that have
,, ~ ~ ~ Am ~ ~ ~ = ~ ~ ~
~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ T ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ i f T ~ W~
—A-— ·~ ~ ~ —·—~ ~ ~ · ~ - - · ~ ~ e ~ - · - ~ ~ ~ ~ * ~ ·~~ a . ~ ~ . .- ., .- , ... ._
. .~.~.~ , ~.~.~.~.~._.~.e,..~.~., ~ , . . ,. , , ., ,._
_., ~ ~.~.~.~.~ ~.~.~.* , ~.~.~ ~ ~.~.~.~_~.~.e,.~.~.
— · - e- -; - - ~ - ·- · - e - - - · ~ - ~ e - *—*— ~ - . - ~ - . - .e ~ .~ . an- _
2 A _-.- ~e ·- -a ~ - " ~ -a " - - 2 _
' ' ' -' 'e' - - - - - ' ' ' - ' ' ' ' - ' ' - ' ' ' ' - - ' ' '.' ' 2 ' ' - ' -2- - 2 ' at' '.' ' ' ~ ' ' .' ' '.' ' ' '.'
. ~ .
"' - ' - - * - - - ~ - - - 2 - - - - - - - ' ' - - ~ - --. - . - -~_. . ' .' .: . '
ma: __: _
a'''-''---'_''-- _
at_ _
. . . . . .
. . . - - - · ~
A.: _._ e
.—,~ .'- ''- - 'a—''"'-'-- s
me: .... :..: ..: ._...:..: .. :...:..: .. :...:. _
_ _ _ :::::
::: :- -: —
'-:--':' 2 ': -:-—Ir
~ , _ _ ~ _ _
C I ~ ~ ~ ~ ~ m
I
m ~ C 0 Cal u'
D 5 ~ m ~ m D i ~ m m D O — ~ ~ 8 ~ ~ s
z ~ m ~ m ~ m ~ ~ ~ ~ ~ ~ N ~ ~ ~ ~ · R z -
~ O
Fig. 4. Crossmap of research fronts to base reference groups for a collection of complex networks papers.
5294 1 www.pnas.org/cgi/doi/10.1073/pnas.0307604100
epidemics
network theory
social networks
food webs
Intemet
epidemics
small worlds
early papers
bio networks
indeterminate
— r
r
O
so ~
~ 6
it ~
Morris and Yen
OCR for page 113
~ — To ~ ~
1 1 1 1 1 1 1 1
· ~ 6 ~ .. ~ ·~ ~ ~ · ~ ~ · - ·
: . i ~ : ~ ~ ~~. ~ ~ ~.~ ·~ ~ ~~ I ~ ~ ~ · ~.~ ~.~ ~ ~~
1 1
< <
N ~
_.
epidemics
net ark theory
social networks
. .
.
. .
1 1
O ~
O z
8 ~ ~ ~ ~ m ~ ~ ~ ~ ~ ~ 3 D = ~ = < ~ ~ o 3
m ~ ~ ~ ~ ~ , w
Fig. 5. Crossmap of research fronts to author collaboration groups for a collection of complex networks papers.
received eight or more citations in the last 12 months. The six
most cited papers are noted in Fig. 3.
The research front labels were added after manually browsing
titles of papers in each research front for themes. In future
research, it may be possible to automate or semiautomate the
labeling process by using correspondence to terms derived from
term co-occurrence groups. When browsing titles, considerable
overlap is found in themes. Note, for example, the label "epi-
demics" is used for both research front 8 and research front 10.
Interestingly, research front 7, "social networks," has many
recent papers that are currently highly cited, implying an im-
portant research topic providing current base knowledge. Re-
search fronts 10 and 8, both labeled epidemics, are both recent,
indicating an emerging topic of research.
Fig. 4 is a crossmap of research fronts to base reference
clusters, which were found by using the cocitation similarity
metric (6~. References cited <20 times were discarded, leaving
50 references for clustering, shown individually. Correspondence
was measured by counting the number of times a reference
appears in papers in a research front. The dendrogram at the top
of the map shows ~8 base reference clusters. The central group,
references 3, 9, and 16 in Fig. 4, are used by all research fronts
except research fronts 4 and 3. Note the difference in the two
epidemics research fronts: research front 10 uses references by
authors Albert and Pastor-Satorras (references 1, 2, and 8),
whereas research front 8 relies heavily on references by authors
Moore, Barrat, and Newman (references 25, 36, and 42~. It is also
easy to see that research fronts 7 and 8 overlap in their use of
references 14-44, but at the same time research fronts 7 and 10
overlap, using references 5-34.
Fig. 5 shows a crossmap of research fronts to author collab-
oration groups. Collaboration groups were found by using
coauthorship counts as the similarity metric. Authors with fewer
than three papers were discarded, leaving 28 authors for clus-
tering, which are individually shown. Correspondence is mea-
sured by counting the number of times an author appears in
1. Borner, K., Chen, C. & Boyack, K. W. (2002) Annul Rev. Ini Sc . Technol. 37,
179-255.
2. White, H. D. & McCain, K. W. (1989) Annul Rev. In; Sci. Technol. 24,
1 19-186.
3. Chen, P. P. S. (1976) ACM Trans. Database Systems 1, 9-36.
4. Small, H. (1978) Social Studies Sci. 8, 327-340.
5. Kessler, M. M. (1963) Am. Doc. 14, 10-25.
6. Small, H. G. (1973) J. Am. Soc. Ini Sci. 24, 265-269.
7. White, H. D. & Griffith, B. C. (1981) J. Am. Soc. In; Sci. 32, 163-171.
8. Subramanyam, K. (1983) J. In; Sci. 6, 33-38.
9. Persson, O. (1994) J. Am. Soc. Ini Sci. Technol. 45, 31-38.
Morris and Yen
food webs
epidemics
small worlds
early papers
bio net orks
indeterminate
papers in a research front. Author groups are easily discerned
from the dendrogram at the top of the map. Example groups are
authors 14-21 (Strogatz, Watts, Newman, and Moore), and
authors 11-2, (Jeong, Albert, and Barabasi). Note the overlap of
author groups 6, 2, and 18 across research fronts 10 and 7.
Additionally, Albert (author 8), whose papers were used as
references from research front 10, does not appear to author any
papers in that research front. Other overlapping relations are
evident, particularly author groups contributing to research
fronts 7 and 9.
Conclusion
The crossmapping technique shown here provides an easily
understood method for exploring relations in a collection of
papers. In the example shown here, two types of crossmaps allow
comprehension of the overlapping relations among research
fronts, base reference groups, and author collaboration groups.
Potential users of crossmaps are researchers exploring the
literature of a scientific field to discover current research topics,
base references, research teams, core journals, and the relations
among them. In our experience, the method is particularly useful
for mapping a domain during the initial state-of-art review phase
of new research projects when the researchers are most unfa-
miliar with a field's literature. The method will be useful for
summarizing information about a field for presentation to
subject matter experts for technology forecasting. User studies
need to be conducted to validate the ease of comprehension of
crossmaps and identify the most useful pairs of entity types for
crossmapp~ng.
The technique has so far been applied mainly to small, well
focused collections of papers (<1,000 papers.) The principal
limitation to crossmapping of large collections of papers is the
restricted space available on the axes for labels. The technique
may be adaptable to very large collections if interactive tools
are added to expand and contract levels of the clustering
hierarchy.
10. Kuhn, T. S. (1970) The Structure of Scientific Revolutions (Univ. of Chicago
Press, Chicago).
11. Morris, S. A., Yen, G., Wu, Z. & Asnake, B. (2003) J. Am. Soc. Inf. Sci. Technol.
5S, 413-422.
12. Small, H. (1997) Scientometrics 38, 275-293.
13. White, H. D. & McCain, K. W. (1998) J. Am. Soc. Inf. Sci. 49, 327-355.
14. Garfield, E. (1979) Citation Indexing: Its Theory and Application in Science,
Technology, and Humanities (Wiley, New York).
15. Kretschmer, H. (1997) Scientometrics 40, 579-591.
16. Crane, D. (1972) Invisible Colleges: Diffusion of Knowledge in Scientific Com-
munities (Univ. of Chicago Press, Chicago).
PNAS 1 April 6, 2004 1 vol. 101 1 suppl. 1 1 5295
OCR for page 114
17. Callon, M., Courtial, J. P. & Laville, F. (1991) Scientometrics 22, 155-205.
18. Lotka, A. J. (1925) J. Wash. Acad. Sci. 16, 317-323.
19. Bradford, S. C. (1938) Engineering 137, 85-86.
20. Zipf, G. K. (1949) Human Behavior and the Principle of Least Effort (Addison-
Wesley, Reading, MA).
21. Naranan, S. (1971) J. Doc. 27, 83-97.
22. Ziegler, E., Kunz, C., Botsch, V. & Schneeberger, J. (2002) in Proceedings
of the IEEE Sixth International Conference on Information Visualization
5296 1 www.pnas.org/cgi/doi/10.1073/pnas.0307604100
(IEEE, Los Alamitos, CA), pp. 361-366.
23. Mothe, J. & Chrisment, C. (2003) J. Am. Soc. Inf. Sci. Technol. 54,
650-659.
24. Shneiderman, B., Feldman, D., Rose, A. & Grau, F. G. (2000) in Fifth ACM
Conference on Digital Libraries (Association for Computing Machinery, New
York), pp. 57-66.
25. Salton, G. (1989)Automatic Text Processing: The Transformation,Analysis, and
Retrieval of Information by Computer (Addison-Wesley, Reading, MA).
Morris and Yen
Representative terms from entire chapter:
entity type
