=Paper=
{{Paper
|id=Vol-1143/paper2
|storemode=property
|title=Systematic Retrieval of Scientific Literature based on Citation Relations: Introducing the CitNetExplorer Tool
|pdfUrl=https://ceur-ws.org/Vol-1143/paper2.pdf
|volume=Vol-1143
|dblpUrl=https://dblp.org/rec/conf/ecir/EckW14
}}
==Systematic Retrieval of Scientific Literature based on Citation Relations: Introducing the CitNetExplorer Tool==
https://ceur-ws.org/Vol-1143/paper2.pdf
Systematic retrieval of scientific literature
based on citation relations:
Introducing the CitNetExplorer tool
Nees Jan van Eck and Ludo Waltman
Centre for Science and Technology Studies, Leiden University, The Netherlands
{ecknjpvan,waltmanlr}@cwts.leidenuniv.nl
Abstract. We consider the problem of scientific literature search, and we sug-
gest that citation relations between publications can be very helpful in the sys-
tematic retrieval of scientific literature. We introduce a new software tool called
CitNetExplorer that can be used for citation-based scientific literature retrieval.
To demonstrate the use of CitNetExplorer, we employ the tool to identify publi-
cations dealing with the topic of community detection in networks. Citation-
based scientific literature retrieval can be especially helpful in situations in
which one needs to obtain a comprehensive overview of the literature on a cer-
tain research topic, for instance in the preparation of a review article.
1 Introduction
Searching for scientific literature can be a time-consuming activity. This is especially
the case when comprehensive search results are needed. For instance, performing a
literature search for a review article that intends to provide a complete overview of all
literature on a certain research topic often takes a lot of time. Likewise, it may require
a substantial amount of effort to systematically search through a body of literature in
order to determine in an accurate way the origins of a certain scientific idea or con-
cept.
In many cases, citation relations between publications are helpful when searching
for scientific literature. For instance, when one has found a relevant publication, addi-
tional relevant publications may be identified by checking the publications cited by
the publication that has already been found. When additional relevant publications
have been found, one may in turn check the publications cited by these publications.
In this way, citation relations may help to identify significant numbers of relevant
publications. Citation relations may also be used in a forward rather than a backward
direction in time. In that case, when one has found a relevant publication, additional
relevant publications may be identified by looking for publications citing the publica-
tion that has already been found.
Although citation relations can be very helpful in the retrieval of scientific litera-
ture, following large numbers of citation links is quite time consuming. To some de-
gree, bibliographic databases such as Web of Science, Scopus, and Google Scholar
can be used to increase the efficiency of citation-based literature retrieval. In Web of
Science and Scopus, for instance, one can easily look up both the publications that
cite and the publications that are cited by a given publication. Nevertheless, biblio-
graphic databases offer only limited support to researchers who want to use citation
relations for systematic literature retrieval. These databases do not give a visual over-
view of the citation relations within a set of publications. (Web of Science does pro-
vide a so-called ‘citation map’ visualization, but the information offered by this visu-
alization is limited.) They also do not support sophisticated citation-based search que-
ries. For instance, when one has available a set of relevant publications, there is no
easy way to identify all publications that cite or are cited by at least three publications
in the set.
In this paper, we introduce a software tool for analyzing and visualizing citation
networks. The tool is called CitNetExplorer and is available at www.citnetexplorer.nl.
We have developed CitNetExplorer primarily for the purpose of studying the evolu-
tion of the scientific literature in a research field [1,2]. However, CitNetExplorer can
also be used for systematic literature retrieval based on citation relations. The tool
may serve as a prototype for citation-based information retrieval functionality that
could be implemented in bibliographic databases such as the ones mentioned above.
The organization of this paper is as follows. In Section 2, we discuss CitNetEx-
plorer in more detail. In Section 3, we demonstrate the use of CitNetExplorer for the
purpose of systematic literature retrieval. Finally, in Section 4, we summarize our
conclusions.
2 CitNetExplorer
CitNetExplorer, which is an abbreviation of ‘citation network explorer’, offers the
following functionality:
Searching for publications based on author, journal, title, etc. This is the traditional
way of searching for scientific literature.
Visualization of the citation network of a set of selected publications.
Identification of sets of related publications by identifying connected components,
clusters, or core sets in a citation network.
Identification of predecessors or successors of a set of selected publications. Prede-
cessors are publications cited by the selected publications. Successors are publica-
tions citing the selected publications. The minimum number of citation relations
required for a publication to be identified as a predecessor or successor is deter-
mined by the user.
Identification of publications located on a citation path between selected publica-
tions. These publications are referred to as intermediate publications.
Drilling down into a citation network. This means reducing the number of selected
publications.
Expanding a citation network. This means increasing the number of selected publi-
cations.
A screenshot of CitNetExplorer is presented in Fig. 1. The user interface of Cit-
NetExplorer consists of a left and a right panel. The right panel shows the selected
publications, either by displaying the citation network of the publications or simply by
displaying the publications in a list. The left panel offers some general information on
the selected publications (i.e., number of publications, number of citation relations,
and time period), it provides access to a number of important parameters related to
CitNetExplorer’s drill down functionality, and it displays detailed bibliographic in-
formation on a single highlighted publication. In the top part of CitNetExplorer’s user
interface, a number of buttons are available. These buttons can be used to perform
some key operations, in particular the drill down and expand operations. They can
also be used to navigate back and forth between different sets of selected publications
and the corresponding citation networks. This is somewhat similar to the back and
forward buttons in a web browser.
Fig. 1. Screenshot of CitNetExplorer.
An example of a citation network visualization produced by CitNetExplorer is
shown in Fig. 2. In this visualization, the circles indicate publications, labeled by the
last name of the first author, and the curved lines indicate citation relations. The verti-
cal dimension represents time, with more recent years being located below earlier
years. The location of a publication in the vertical dimension is determined by the
year in which the publication appeared. In the horizontal dimension, publications are
located close to each other if they are closely connected in the citation network. Pub-
lications that do not have a close connection in the citation network are located further
away from each other. Similar to software tools for exploring geographical maps (e.g.,
Google Maps) and also to our own VOSviewer tool [3], CitNetExplorer offers zoom-
ing and panning (scrolling) functionality. This can be used to explore specific areas in
the visualization of a citation network in more detail. CitNetExplorer can handle very
large citation networks, with millions of publications and tens of millions of citation
relations. However, in the visualization of a citation network, only a selection of all
publications are displayed. By default, the 40 most frequently cited publications in a
citation network are included in a visualization.
The idea of a software tool for visualizing citation networks of scientific publica-
tions is not new. This idea can also be found in the HistCite tool for ‘algorithmic his-
toriography’ developed by Eugene Garfield and colleagues [4] (see
www.histcite.com). However, in addition to visualizations of citation networks, Cit-
NetExplorer also offers extensive functionality for exploring and analyzing citation
networks. Given the large size of the citation networks that can be handled by Cit-
NetExplorer, a crucial element of CitNetExplorer is the functionality for drilling down
into a citation network. This functionality for instance makes it possible to start at the
level of a full citation network consisting of several millions of publications and to
then gradually drill down into this network until a small subnetwork has been reached
including no more than, say, 100 publications, all dealing with a specific topic of
interest. While drilling down reduces the number of publications that are being con-
sidered, expansion has the opposite effect and increases the number of publications
under consideration. When one finds oneself in a given subnetwork, CitNetExplorer
makes it possible to expand the subnetwork by adding publications cited by or citing
to one or more publications in the subnetwork. In addition to drill down and expand
functionality, CitNetExplorer also provides a number of options for analyzing citation
networks. Examples include options for clustering the publications in a citation net-
work and for identifying the core publications in a citation network. As we will
demonstrate in the next section, CitNetExplorer’s functionality for exploring and
analyzing citation networks enables a more sophisticated approach to citation-based
retrieval of scientific literature compared with the possibilities provided by biblio-
graphic databases such as Web of Science, Scopus, and Google Scholar.
CitNetExplorer has been implemented in Java and therefore works on any system
that offers Java support. The tool can be downloaded from www.citnetexplorer.nl.
Ideally, one would like a tool such as CitNetExplorer to be linked directly to a biblio-
graphic database. Although this is technically feasible, the proprietary nature of bibli-
ographic databases does not allow us to create such a direct link. Users of CitNetEx-
plorer therefore need to take care themselves of obtaining the publication and citation
data required by CitNetExplorer. They can then use their data as input to CitNetEx-
plorer. To support users in these steps, CitNetExplorer is able to directly process data
that has been downloaded through the web interface of Web of Science. However,
because of restrictions imposed by the Web of Science web interface (i.e., data can be
downloaded for at most 500 publications at a time), only small and medium-sized
citation networks can be handled in this way. To work with large citation networks,
for instance including several millions of publications, one needs to have direct and
unrestricted access to Web of Science or another bibliographic database. Most re-
searchers do not have this type of access to a bibliographic database. In that case, the
use of CitNetExplorer is limited to small and medium-sized citation networks, per-
haps including at most several tens of thousands of publications. Of course, even with
this limitation, CitNetExplorer can still serve as a prototype for citation-based infor-
mation retrieval functionality that could be implemented at a large scale in biblio-
graphic databases.
3 Demonstration
To demonstrate the use of CitNetExplorer for citation-based retrieval of scientific
literature, we take as an example the topic of community detection in networks (for a
review of the literature on this topic, see [5]). We use this topic because we are famil-
iar with it and because we are therefore able to assess which publications are relevant
and which are not. Our focus is on the community detection literature in the field of
physics. We do not consider publications from other fields, but we do take into ac-
count publications in multidisciplinary journals such as Nature and Science.
For the purpose of this demonstration, CitNetExplorer receives as input a citation
network of publications in physics journals and multidisciplinary journals in the peri-
od 1998–2012. The data is taken from our institute’s in-house version of the Web of
Science database. The citation network includes about 1.8 million publications and
about 15.1 million citation relations.
To retrieve in a systematic way the literature on community detection, we take the
following five steps in CitNetExplorer:
1. First, we search for all publications whose title matches ‘*communit* detect*’ or
‘*detect* communit*’. This yields 113 publications. We drill down to the subnet-
work consisting of these 113 publications.
2. Some of the publications in our subnetwork are false positives and do not deal with
community detection. An example is a publication with the title ‘Detection of large
numbers of novel sequences in the metatranscriptomes of complex marine micro-
bial communities’. We assume that false positives have no citation relations with
publications that have been correctly identified as being relevant to the topic of
community detection. We therefore identify the largest connected component in
our subnetwork and remove from the subnetwork the seven publications not in-
cluded in this component.
3. We now have a subnetwork consisting of 106 publications. The visualization of
this subnetwork produced by CitNetExplorer is shown in Fig. 2. Recall from the
previous section that only the 40 most frequently cited publications are included in
the visualization. We observe that some important early publications on communi-
ty detection are not included in our subnetwork. We therefore expand the subnet-
work with 33 publications that are each cited by at least ten publications already
included in the subnetwork.
4. Within our subnetwork of 139 publications, some of the early publications deal
with networks but not specifically with community detection. An example is a pub-
lication with the title ‘Statistical mechanics of complex networks’. We manually
identify six publications not dealing specifically with community detection, and we
remove these publications from the subnetwork.
5. Finally, we expand our subnetwork of 133 publications with 439 publications that
each cite at least four publications already included in the subnetwork. A visualiza-
tion of the subnetwork that is obtained after the expansion is presented in Fig. 3.
Fig. 2. Visualization of a citation network of 106 publications on the topic of community detec-
tion in networks. Only the 40 most frequently cited publications are included in the visualiza-
tion. To reduce the number of lines in the visualization, some citation relations are not dis-
played. For instance, if publication A cites both publication B and publication C and if publica-
tion B also cites publication C, then the citation relation from publication A to publication C is
not displayed. Only citations relations included in the so-called transitive reduction of the cita-
tion network are displayed.
After taking the above five steps, we end up with a set of 572 publications. The
large majority of these publications turn out to deal with community detection, alt-
hough some less relevant publications are included as well. These are for instance
review articles that cite a number of publications on community detection even
though this topic is not their main focus. Some further manual work would be needed
to get rid of these less relevant publications. What is important is that we may expect
a very large share of all publications dealing with community detection to be included
in our set of 572 publications. Relevant publications may have been missed only if
they cite no more than three publications out of the 133 publications that we start with
in step 5. Since these 133 publications include a number of the most important publi-
cations on community detection, this does not seem very likely.
Fig. 3. Visualization of a citation network of 572 publications on the topic of community detec-
tion in networks. Only the 40 most frequently cited publications are included in the visualiza-
tion. Like in Fig. 2, to reduce the number of lines in the visualization, only citations relations
included in the so-called transitive reduction of the citation network are displayed.
4 Conclusions
We have discussed the idea of using citation relations between publications to support
the systematic retrieval of scientific literature. We have introduced the CitNetExplorer
software tool that can be used for this type of citation-based information retrieval. We
believe that our approach is especially helpful in situations in which one needs to get
a comprehensive overview of the literature on a certain research topic, for instance
when working on a review article or when tracing the origins of a scientific idea or
concept. Although our focus has been on the retrieval of scientific literature, we note
that our approach may also be useful in the retrieval of patents.
The retrieval performance of our approach has not yet been evaluated in a formal
way, for instance based on precision and recall measures. However, during the devel-
opment of the CitNetExplorer tool, we have used the tool extensively to test its use-
fulness in systematic literature search. We consider the results of these tests to be
encouraging. In a number of cases, for instance, CitNetExplorer has been helpful in
the identification of interesting publications that so far we had not been aware of. On
the other hand, one should of course keep in mind the limitations of the use of citation
relations in scientific literature retrieval. Obviously, when two scientific communities
both work on the same topic but do not cite each other’s publications, the use of cita-
tion relations will be of no help in identifying the connection between the work of the
two communities.
Finally, we briefly mention some related work. There are at least two areas in
which related work can be found. On the one hand, there is the area of information
visualization and human-computer interaction. Examples of related work in this area
include work on the visualization of citation networks to support literature reviewing
[6] and work on the combined use of citation network visualization and natural lan-
guage processing to support researchers in the exploration of a body of scientific liter-
ature [7]. On the other hand, there is the area of information retrieval. In this area,
researchers study the problem of citation recommendation (e.g., [8]). This is about the
situation in which someone is working on a publication and is looking for related
publications to cite. These related publications can be identified based on textual simi-
larity or based on citation relations, and especially in the latter case there is a relation-
ship to our work.
References
1. Van Eck, N.J., Waltman, L.: Visualizing bibliometric networks. Manuscript in preparation
(2014)
2. Van Eck, N.J., Waltman, L.: CitNetExplorer: A new software tool for analyzing and visu-
alizing citation networks. Manuscript in preparation (2014)
3. Van Eck, N.J., Waltman, L.: Software survey: VOSviewer, a computer program for bibli-
ometric mapping. Scientometrics 84, 523–538 (2010)
4. Garfield, E., Pudovkin, A.I., Istomin, V.S.: Why do we need algorithmic historiography?
Journal of the American Society for Information Science and Technology 54, 400–412
(2003)
5. Fortunato, S.: Community detection in graphs. Physics Reports 486, 75–174 (2010)
6. Chou, J.-K., Yang, C.-K.: PaperVis: Literature review made easy. Computer Graphics Fo-
rum 30, 721–730 (2011)
7. Dunne, C., Shneiderman, B., Gove, R., Klavans, J., Dorr, B.: Rapid understanding of sci-
entific paper collections: Integrating statistics, text analytics, and visualization. Journal of
the American Society for Information Science and Technology 63, 2351–2369 (2012)
8. Strohman, T., Croft, W.B., Jensen, D.: Recommending citations for academic papers. In:
30th Annual International ACM SIGIR Conference, pp. 705–706. ACM, New York (2007)