Decomposing Social and Semantic Networks in Emerging “Big Data” Research

http://www.sciencedirect.com/science/article/pii/S1751157713000473
빅데이터가 학문으로 등장한 배경을 잘 정리한 논문
http://www.sciencedirect.com/science/article/pii/S1751157713000473
Park, H.W., & Leydesdorff, L. (2013). Decomposing Social and Semantic Networks in
Emerging “Big Data” Research. Journal of Informetrics. 7 (3), 756-765. DOI
information: 10.1016/j.joi.2013.05.004
Decomposing Social and Semantic Networks in
Emerging “Big Data” Research
Han Woo Park 1
Department of Media and Communication
Yeungnam University, South Korea
Loet Leydesdorff
Amsterdam School of Communication Research (ASCoR), University of Amsterdam
Kloveniersburgwal 48, 1012 CX Amsterdam, The Netherlands (loet@leydesdorff.net)
<Author Biography>
Han Woo Park (Corresponding Author) is Full Professor in the Department of Media and
Communication at Yeungnam University, South Korea. He is also the director of the
CyberEmotions Research Institute at Yeungnam University. He was the principal investigator
of the World Class University (WCU) Webometrics project (2009-2011). His research is
available at http://www.hanpark.net.
Loet Leydesdorff (Ph.D., Sociology; M.A., Philosophy; M.Sc., Biochemistry) is a Professor
at the Amsterdam School of Communication Research (ASCoR) of the University of
1
Corresponding Author: Department of Media and Communication, YeungNam University, 214-1, Dae-dong,
Gyeongsan-si, Gyeongsangbuk-do, 712-749, South Korea, E-mail: hanpark@ynu.ac.kr,
parkhanwoo@hotmail.com; Telephone: 82-53-810-2275; Fax: 82-53-810-2270).
1

Amsterdam. He is a Visiting Professor at the Institute of Scientific and Technical Information
of China (ISTIC) in Beijing and an Honorary Professor at the Science and Technology Policy
Research Unit (SPRU) of the University of Sussex. He has published extensively in systems
theory, social network analysis, scientometrics, and the sociology of innovation (for his
works, see http://www.leydesdorff.net/list.htm).
<Acknowledgements>
The authors would like to thank Seong-Cheol Choi, Ji-Young Kim, Ji-Young Park, Yun-
Cheol Heo, and Min-Ho So for their assistance during the collection and organization of data
and acknowledge support from the SSK (Social Science Korea) Program of the National
Research Foundation of Korea (NRF-2010-330-B00232).
Abstract
This paper examines the structural patterns of networks of internationally co-authored SCI
papers in the domain of research driven by big data and provides an empirical analysis of
semantic patterns of paper titles. The results based on data collected from the DVD version of
the 2011 SCI database identify the U.S. as the most central country, followed by the U.K.,
Germany, France, Italy, Australia, the Netherlands, Canada, and Spain, in that order.
However, some countries (e.g., Portugal) with low degree centrality occupied relatively
central positions in terms of betweenness centrality. The results of the semantic network
analysis suggest that internationally co-authored papers tend to focus on primary
technologies, particularly in terms of programming and related database issues. The results
show that a combination of words and locations can provide a richer representation of an
emerging field of science than the sum of the two separate representations.
Keywords: Data Science, Big Data, International Co-authorship, Social Network Analysis,
SCI, Semantic Network, Scientometrics
Highlights
2

► This study conducts a social network analysis of international co-authorship relationships
in big data science and analyzes semantic networks ► The results indicate that the U.S. was
the most central country, followed by Germany, the U.K., and France, in that order. ► Some
European countries engaged in international collaboration to an extent greater than expected.
3

Introduction
There has been growing interest among scientists, policymakers, industry
professionals, and journalists in the development of research driven by big data as
technologies and practices for research and development worldwide (Halevi & Moed, 2012;
OECD, 2012; O’Reilly Radar Team, 2012; Rousseau, 2012). The term “big data” refers to
“analytical technologies that have existed for years but can now be applied faster, on a greater
scale and are accessible to more users.” (Miller, 2013, p. 1). In this regard, research driven
by big data reflects a discipline that, to extract meaning from very large datasets, incorporates
various techniques such as data mining and visualization into diverse fields, including the
humanities and social sciences. In a special issue published on February 11, 2011, Science
looked broadly at research efforts driven increasingly by big data as a scientific domain
(Science, 2011). Several prominent institutions have organized international conferences, for
example, “IPP 2012: Big Data, Big Challenges” conference by the University of Oxford
(http://blogs.oii.ox.ac.uk/policy/?p=447).
Because the most noticeable thread running through national competitiveness in an
increasingly digital society is to cope with the scalability of data, research driven by big data
has become a priority in both academia (Harris, 2012; King, 2011) and industries (McKinsey
Global Institute, 2011; Rogers, 2011). Because big data can have substantial impact on
various aspects of society, the “Big Data Public Private Forum” was also established in order
to offer a major boost for supporting actions for the successful implementation of “the big
data economy” (http://www.big-project.eu/). According to an EMC report on global big data
(http://www.emc.com/about/news/press/2012/20121211-01.htm), Western Europe is currently
making the heaviest investment in managing the deluge of digital data, spending $2.49 per
gigabyte (GB), followed by the U.S. ($1.77), China ($1.31), and India ($0.87).
This study investigates the visibility of countries in terms of their research
collaboration in the domain of big data in the context of scientific documents. The study
examines the use of related terms by country based on social and semantic network methods
and provides a quantitative analysis of how various terms are communicated in academia.
More specifically, the study provides an empirical analysis using SCI (Science Citation
Index) journals for the year 2011. We consider the following research questions:
4

RQ1) What is the structural pattern of the network of internationally co-authored SCI
papers in the domain of research driven by big data? Which countries occupy central or
peripheral positions in the 2011 network?
RQ2) What is the semantic structure of paper titles in research driven by big data?
Literature Review
This section provides a theoretical introduction based on a literature review. Big data
research is composed of interrelated clusters of tasks. For example, technologies for
accessing, collecting, and organizing large amounts of data and unique skill sets have played
an increasingly central role in big data research (Loukides, 2012). Rousseau (2012) claimed
that the definition of “large amounts of data,” i.e., big data, depends on concrete
circumstances related to individual projects. It may mean tens of terabytes (1012
≈ 240
bytes)
for some research, whereas it may mean several petabytes (1015
≈ 250
bytes) or even exabytes
(1018
≈ 260
bytes) for large enterprises and huge research projects. From this perspective,
Loukides (2012) claimed that today’s “big” is probably tomorrow’s “medium” and next
week’s “small” and thus that the most effective definition of “big data” may be derived when
the size of data itself becomes part of the research problem.
The origin of big data research is rooted mainly in various initiatives started in the
early 1970s. One such initiative includes Peter Naur’s 1974 book “Concise Survey of
Computer Methods,” a survey of contemporary data-processing methods in a wide range of
applications (Gilpress, 2012). During the 1990s, the term was associated mainly with
computer modeling and software development for large data sets. The launch of the journal
Knowledge Discovery and Data Mining in 1997 represents such a trend in computer science.
Rousseau (2012) regarded the early 1990s as the beginning of the field of big data research.
Recent years have witnessed the emergence of scholarly journals focusing mainly on research
driven by big data, including Data Science Journal (2002), Journal of Data Science (2003),
EPJ Data Science (2012), and Journal of Big Data (2013). While these journals address the
general aspects of big data research, GigaScience, established in 2012, focuses specifically on
the study of “big data” from the entire spectrum of life and biomedical sciences.
Although research based on big data can be conducted in various ways, its basic
purpose lies in handling huge amounts of data from technological, sociological, and
5

economic systems to discover some hidden patterns. This is referred to as the “fourth
paradigm,” which has been described as a new approach to scientific exploration. This
scientific approach is attributed to Jim Gray, one of the most influential computer scientists at
Microsoft (Hey, Tansley & Tolle, 2009). The establishment of big data research as a
complementary approach to traditional hypothesis-driven methods started in natural and
engineering domains and later moved to social and behavioral sectors in the mid-2000s
(Bonacich, 2004). Savage and Burrows (2007) stated that big data research is closely
associated with changes in conducting scientific tasks and lamented that “fifty years ago,
academic social scientists might be seen as occupying the apex of the – generally limited –
social science research ‘apparatus’. Now they occupy an increasingly marginal position in the
huge research infrastructure.” (p. 886).
In particular, the rapid development of mapping-oriented visualization and e-research
practices, together with the universality of advanced research networks and cloud computing,
has facilitated the dissemination of big data research in the late 2000s (Jankowski, 2009). The
volume of data available for research has suddenly become controversial since the “end of
theory” claim by Anderson (2008), who argued that data are everything researchers need and
thus that they do not have to settle for models. Knowledge can be discovered through
correlations (instead of causality) between big data and related events/observations,
experiments, and simulations by enriching traditional data in novel ways, posing a serious
challenge to social scientists. In response, Graham (2012) suggested that “We may one day
get to the point where sufficient quantities of big data can be harvested to answer all of the
social questions that most concern us. I doubt it though. There will always be digital divides;
always be uneven data shadows; and always be biases in how information and technology are
used and produced.” Both the definition and implementation of big data research may still
need to be clarified for its successful adoption in a society. However, several researchers have
become increasingly concerned with big gaps of knowledge in the changing research ecology
made possible by big data (Snijders, Matzat, & Reips, 2012). The creation, organization, and
analysis of big data can be shaped by a number of interdependent social, cultural, and
political subsystems in the research ecosystem. To reduce the barrier to entry into big data
research, the OECD initially suggested an international framework in the following five
domains for open access to scholarly dataware houses ranging from gigabytes to petabytes of
data (Arzberger et al., 2004a, 2004b): technological, institutional/managerial,
6

financial/budgetary, legal/policy, and cultural/behavioral domains. The OECD (2012)
recently organized a technology foresight forum in Paris (“Big Data Analytics and Policies”)
to harness big data as a new source of growth. In the Europe 2020 strategy, the European
Commission identified the use of big data as a new driver of growth and jobs under seven
flagship initiatives (http://ec.europa.eu/europe2020).
Because data streams from many sources can be overwhelming, the ubiquitous use of
big data across various segments of society does not necessary mean intellectual coherence.
Halevi and Moed (2012) examined the development of research related to big data by using
the Scopus database. They searched for the term “big data” by using the index and author
keyword fields from 1970 to early 2012. An examination of 306 documents revealed a sharp
increase in such research in conference proceedings (131 papers), particularly in computer
science (171 documents). In a similar vein, Rousseau (2012) collected 141 documents using
the web version of the Web of Science database based on TS (topic search)=“big data” and
found the salience of conference proceedings (54 papers vs. 47 articles) in computer-related
fields from 1993 to October 2012. Both studies concluded that this research was led mainly
by computer science in the early years (until 2006) but spread to other engineering and
natural science fields in recent years. Further, the U.S., China, and Germany were the most
active countries in terms of the publication of papers based on big data.
Previous studies have not systematically examined whether research efforts driven by
various sources of big data are really becoming increasingly widespread across the world.
Further, the status of the literature based on big data has not been extensively discussed or
sufficiently examined with respect to its semantic variations, disciplinary scope, institutional
adoption, and international collaboration. For example, Halevi and Moed (2012) and
Rousseau (2012) focused on descriptive statistics. Therefore, the present study incorporates a
network perspective into both social (in terms of co-authorship) and semantic networks. In
addition, the study extends search queries to various terminologies related to research
enabling big data because the term “big data” is regarded as only one of many terms. The
study shows where the research system focusing on big data is particularly popular in terms
of international collaboration and prevailing semantics.
Methods: Data Collection and Analysis Procedures
7

Social Network Analysis and the Co-Authorship Network
We employed a method rooted in social network analysis (SNA) (for a detailed
explanation of social network metrics and technical procedures, see Hanneman & Riddle,
2005). Here the unit of analysis is often the node, which refers to a point in a network where
ties cross or connect nodes. A tie is a connection between parts (i.e., nodes) in a network. In
the present study, we considered countries as nodes and a tie as the number of papers co-
authored by a pair of researchers with different addresses in terms of their country of origin.
Previous studies (Kwon, Park, So & Leydesdorff, 2012; Otte & Rousseau, 2002; Park &
Leydesdorff, 2008, 2010; Shapiro & Park, 2012) have shown that indicators and
visualizations based on an SNA are particularly useful for identifying the structural pattern of
research collaboration between a set of authors. In addition, a “new invisible college”
embedded in scientific networks can be disclosed (Wagner, 2008).
Data Collection
For the data, we considered papers published in SCI journals in 2011. For this, we selected
three types of documents: journal articles, letters, and reviews. We obtained the data from the
DVD version of the SCI database by using several search terms based on titles, author
keywords, and keyword-plus. In addition, we considered the space between two words. For
example, we used the terms “big data” and “bigdata” as a search string.
We made search strings based on the following procedure: First, we attempted to find
some authoritative sources listing all relevant keywords, but to the authors’ knowledge, no
related studies provide a solid list of queries. Therefore, we required a more objective
technique. Second, we used the term “big data” as a seed search string for “titles,” “author
keywords,” and “keyword plus” to retrieve relevant documents from the DVD version of the
SCI 2011 database. However, we retrieved only four documents with the term “big data.” We
considered including some terms co-occurring with “big data,” including
“cyberinfrastructure,” “data mining,” and “clouds.” To complement the SCI DVD, we also
considered the SCI(E) 2011 database from the Web of Science (WoS) database by using
Topic=("Big Data") as the search string and retrieved a total of 15 documents. The important
terms related to “big data” included “data integration” and “computational science.” Third,
8

the term “big data” in the title provided 889 documents through Google Scholar. We included
the most frequently co-occurring title words with “big data” in this set, including “clouds,”
“cloud computing,” “Hadoop,” “MapReduce,” “big science,” “NoSQL,” and “data
warehouse.” Therefore, we used a total of 11 keywords (“big data,” “big science,” “cloud
computing,” “computational science,” “cyberinfrastructure,” “data integration,” “data
mining,” “data warehouse,” “Hadoop,” “MapReduce,” and “NoSQL”) for the final search
string.
Before the finalization of search queries, we personally sent a list of potential queries
to Rousseau (personal communication, February 27, 2013), who provided a scientometric
mapping of the development of the domain of big data research by using the web version of
the bibliometric database (Rousseau, 2012). During the consultation, we realized that the
term "parallel computing" has little to do with big data research because this technique has
been studied and applied in the last few decades. Finally, we removed duplicate documents
collected from a total of 477 records by using 11 search strings and thus had a total of 406
records for the final analysis.
Analysis Methods and Indicators
Using a list of scientific documents selected as relevant to big data science, we
created a binary matrix containing 406 records of documents (articles, reviews, and letters) as
rows and countries as variables (i.e., columns). Here if one country has four addresses listed
and another country, two, then some social network analysis tools count them as 4 x 2 = 8
collaborations. In the case of binary values, there is a single (1 x 1) collaboration between
these two countries. Given the research questions, we used this binary file. We considered as
missing records those with no addresses.
We used this asymmetric binary matrix in UCINet to generate a symmetrical and
valued co-authorship network. We first examined the structural properties of this co-
authorship network in the field of big data science relative to those of the global co-
authorship network in 2011 (Leydesdorff, Wagner, Park & Adams, 2013). After the
normalization, we used the global distribution to specify expected numbers for a more in-
depth analysis.
9

Structural properties can be detected in several ways. A major indicator is network
density, which is the total number of ties divided by the total number of possible ties. For a
valued network, this index is the total of all values divided by the number of possible ties
(Borgatti, Everett & Freeman, 2002). Along with density, centralization is a useful measure
for describing how densely the ties are concentrated around a focal node in a given network.
This is expressed as a percentage with respect to the most central or unequal network possible
for any number of nodes (Hanneman & Riddle, 2005). For example, in a “star” network,
there is only one node connected to all other nodes in the network. Therefore, network
centrality is 100% because in such a network, all nodes are connected only to the central
node.
To calculate the correlations between the global network and its subset limited to big
data science, we employed the quadratic assignment procedure (QAP) in UCINet. The
Pearson correlation in the QAP is regarded as a standard measure when both networks have
valued (non-binary) relationships measured at the interval level (Hanneman & Riddle, 2005).
Here a low probability (p< 0.05) suggests a fairly strong relationship between matrices that is
unlikely to have occurred by chance.
In addition, we statistically tested the international co-authorship matrix for any
significant differences from the expectations (based on the whole database). As suggested by
Leydesdorff and Bornmann (2012), the statistical significance of performance differences
between two proportions can be tested using the z-test for independent proportions. An
absolute value of z greater than 1.96 indicates a significance difference between two networks
at the 5% level (p < 0.05). Here the critical values for a test at the 1% level (p < 0.01) and the
0.1% level are |z| > 2.576 and greater than 3.29, respectively
(http://easycalculation.com/statistics/p-value-for-z-score.php). For a more rigorous
conclusion, we used the most demanding significance level (p < 0.001).
We considered degree centrality to measure the number of countries with which a
country engaged in direct co-authorship collaboration (Park & Leydesdorff, 2008). Here the
more a node (in this study, a country) is connected directly to other nodes in the network, the
more central the node is (for a detailed calculation of network metrics, see Wasserman &
Faust, 1994). This means that researchers in a central country are more likely to have
opportunities for co-authorship collaboration (Hanneman & Riddle, 2005). We used degree
centrality to measure the direct relationship between a country and its neighbors in a co-
10

authorship network (e.g., reputation, popularity, activity, and prominence in the flow of
information).
In addition to degree centrality, betweenness centrality measures the strategic position
of the country in terms of coordinating scientific collaboration between countries
(Leydesdorff, 2007; Park, 2012). There are two major indicators of betweenness centrality:
node and flow betwenness. The node betweenness centrality of a country refers to the
proportion of all geodesic relationships between pairs of other nodes (in this study, countries)
that pass through a given node (Freeman, 1979). In other words, betweenness centrality refers
to the extent to which a node lies along the shortest path connecting others in the network.
Freeman, Borgatti, and White (1991) expanded betweenness centrality by considering the
number of ties (i.e., the strength of ties connecting two nodes in a network) between pairs of
other nodes that pass through a node for some maximum flow. For a node, flow betweenness
centrality sums up how involved that node is in all the flow between all other node pairs
(Hanneman & Riddle, 2005). Because an international co-authorship network is a valued
symmetric graph only with integer values, flow betweenness centrality can thus be a good
measure of a node’s contribution to all possible maximum flows (Borgatti, Everett &
Freeman, 2002).
Finally, we calculated “structural holes” (Burt, 1992) to estimate the strategic
management of a country in co-authorship networks. From Burt’s perspective, one must
form a tie with another that has strong relationships with others with whom one does not yet
have ties. Selecting such a collaborator reduces communication redundancy and increases
networking efficiency in terms of one’s own ego network. In the analysis of co-authorship
networks in information and library science fields, Abbasi, Chung, and Hossain (2012)
associated a researcher’s structural hole with their scholarly performance and found that
researchers who engage in strong collaboration with one primary co-author within a group of
researchers for joint publication perform better than those with many overlapping
relationships within the same group of co-authors. We calculated three measures of Burt’s
structural holes (Burt, 1992). “Network effectiveness” measures the effective size of the ego
network. In this case, effective size denotes the number of non-redundant countries in the
international co-authorship network. “Network efficiency” is derived by dividing effective
size by the number of countries in the ego network. “Network constraints” measure the extent
to which one’s opportunities (e.g. time and resources that can be used for forming new ties
11

and seeking information) are limited by being connected to others who are already connected
to one another. We employed UCINet mainly for social network measures (e.g., density,
centrality, and sub-clusters) for the network (Borgatti, Everett & Freeman, 2002).
Results
RQ 1 - International Co-Authorship Network
We included a total of 201 countries in the 2011 co-authorship network for the
analysis of big data science. We aggregated addresses in England, Scotland, Wales, and
Northern Ireland as U.K. addresses. Leydesdorff, Wagner, Park, and Adams (2013) provided
a detailed discussion on the whole structure of the 2011 co-authorship network at the global
level. This section discusses the changes in the network structure across different matrices
(Table 2). As expected, the global co-authorship network was far denser than the subnetwork,
that is, co-authorship in big data research. Note that these were not really co-authorship
relationships between countries but relationships between them measured in terms of co-
authorship relationships. The sum of ties in the global network and that of the subnetwork
were 1,073,764 and 10,798, respectively. In addition, the global network was more
centralized around hub countries than the network of big data science in terms of all three
measures of centrality. However, the QAP correlation between the whole 2011 co-authorship
network and big data research demonstrates their significant relationship: this (Pearson)
correlation was .740 (p < .001).
Table 2. A Comparison of Density and Centralization Values
Network
Type
Density (S.D.) Centralization (%)
Degree Node Flow
Global 26.71 (245.70) 5.11 10.08 9.83
Big Data 0.01 (0.18) 4.37 2.70 2.28
N=201.
To determine the difference between the big data science network and the grand
matrix of all co-authorship relationships, we conducted a z-test. The results show that some
countries were overrepresented in the co-authorship network in data science, whereas others,
12

underrepresented. As shown in Table 3, a total of 30 countries had a test value greater than |
1.96| (p<0.05). Among these, 22 had positive z-scores, indicating that group 1 (in this case,
the co-authorship network in big data science) showed higher frequency than group 2 (in this
case, the whole co-authorship network). Among these 22 countries, 50.0% (11) showed z-
scores greater than |3.29| (p<0.001), and 27.3% (6), greater than |2.576| (p<0.01). Macedonia
showed the highest z-scores, but this may be due to a low expected value. Contrary to
expectations, two of Macedonia’s 205 international collaborations were in the set of big data
science. Several European countries (Denmark, Ireland, Portugal, Norway, and the
Netherlands) engaged in international collaboration more than they were expected to (p<.05).
This was likely due to an E.U. policy, but the results provide no clear evidence.
There were some developing and transitional countries, including Taiwan, Singapore,
and South Korea, in the group that scored higher than expected. On the other hand, a total of
six countries had negative z-scores. None of these countries had z-scores less than 3.29
(p<0.001); three, less than 2.576 (p<0.01); and five, less than 1.96 (p<0.05). Another
noteworthy result is that Russia, India, and Japan were underrepresented in international
collaboration in big data science at the 5% level. Wagner, Leydesdorff, Adams, and Park
(2013) provided a detailed discussion on the 2011 global co-authorship network in terms of
science and technology policies.
Degree and Betweenness Centrality
The U.S. was the most central country (Table 4) in terms of degree centrality, acting
as the network’s hub in research involving big data. The U.S. had the largest number of
papers co-authored with other countries, and its degree centrality was 4.450, accounting for
17.5% of the network, followed by Germany, the U.K., France, Australia, the Netherlands,
China, Denmark, and Canada, in that order. In addition to China, some other countries in Asia
were part of the top 20 collaborators, including Taiwan and South Korea. Table 4 lists the top
20 countries in terms of their nodal and flow betweenness. Canada, Italy, South Korea, and
Japan performed better than their degree centrality. Therefore, these countries, with their high
betweenness centrality, exerted control over scientific communication as broker or
gatekeepers.
13

Table 4. Centrality Values for Countries
Rank Country Degree Rank Country Betweenness Rank Country FlowBet
1 U.S. 4.450 1 U.S. 2.734 1 USA 2.309
2 GERMANY 1.650 2 FRANCE 1.253 2 FRANCE 0.929
3 U.K. 1.600 3 U.K. 0.680 3 CANADA 0.537
4 FRANCE 1.400 4 CANADA 0.643 4 ITALY 0.510
5 AUSTRALIA 1.150 5 ITALY 0.620 5 UK 0.377
6 NETHERLAND 1.150 6 AUSTRALIA 0.602 6 SOUTH_KORE 0.359
7 CHINA 1.100 7 SOUTH_KORE 0.346 7 BELGIUM 0.331
8 DENMARK 0.950 8 GERMANY 0.291 8 AUSTRALIA 0.328
9 CANADA 0.900 9 BELGIUM 0.290 9 JAPAN 0.262
10 TAIWAN 0.850 10 PORTUGAL 0.266 10 SLOVENIA 0.200
11 ISRAEL 0.750 11 JAPAN 0.256 11 PORTUGAL 0.185
12 SOUTH_KORE 0.750 12 CHINA 0.137 12 CHINA 0.132
13 SWEDEN 0.750 13 NETHERLAND 0.104 13 SPAIN 0.129
14 ITALY 0.700 14 DENMARK 0.099 14 GERMANY 0.108
15 PORTUGAL 0.700 15 SAUDI_ARAB 0.088 15 MALAYSIA 0.103
16 IRELAND 0.650 16 SLOVENIA 0.068 16 TANZANIA 0.095
17 NORWAY 0.650 17 TAIWAN 0.057 17 VENEZUELA 0.095
18 SPAIN 0.650 18 SPAIN 0.055 18 NETHERLAND 0.089
19 SINGAPORE 0.500 19 ISRAEL 0.037 19 SAUDI_ARAB 0.071
20 SWITZERLAN 0.450 20 AUSTRIA 0.036 20 AUSTRIA 0.063
Structural Holes
Table 5 summarizes some measures of the “structural hole.” A closer examination of
structural holes reveals noteworthy results for the top 20 countries across the three measures.
Central countries in terms of their degree and betweenness centrality also occupied central
positions in terms of network effectiveness. However, a majority of these countries are not
listed in terms of their network efficiency. In addition, European countries rank high in terms
of the constraint indicator. Unexpectedly, the U.S. showed very low efficiency (0.361, 51st
)
and moderately low constraint (0.573, 16th
) indices. This suggests that U.S. researchers may
not be strategically positioned for generating new ideas and enjoying greater career mobility
at the international level and perhaps that the domestic market is more important for these
scholars.
Table 5. Structural Hole Values by Country
14

Rank Country Effectiveness Rank Country Efficiency Rank Country Constrain
1 U.K. 13.071 1 EGYPT 1.000 1 DENMARK 0.312
2 AUSTRALIA 12.879 2 INDIA 1.000 2 NETHERLAND 0.331
3 FRANCE 12.562 3 POLAND 1.000 3 PORTUGAL 0.338
4 U.S. 11.563 4 UZBEKISTAN 1.000 4 ISRAEL 0.343
5 GERMANY 10.746 5 GREECE 0.805 5 NORWAY 0.345
6 NETHERLAND 8.873 6 JAPAN 0.789 6 IRELAND 0.352
7 DENMARK 8.530 7 AUSTRIA 0.725 7 UK 0.364
8 PORTUGAL 8.229 8 BRAZIL 0.722 8 SWEDEN 0.365
9 ISRAEL 8.208 9 NEW_ZEALAN 0.722 9 AUSTRALIA 0.381
10 CANADA 7.672 10 MALAYSIA 0.698 10 GERMANY 0.397
11 ITALY 7.554 11 AUSTRALIA 0.678 11 FRANCE 0.411
12 IRELAND 7.252 12 SAUDI_ARAB 0.667 12 CANADA 0.532
13 NORWAY 7.214 13 IRAN 0.667 13 ITALY 0.535
14 SOUTH_KORE 6.365 14 THAILAND 0.667 14 SAUDI_ARAB 0.548
15 CHINA 6.057 15 SINGAPORE 0.659 15 SWITZERLAN 0.556
16 SWEDEN 5.978 16 CZECH_REPU 0.644 16 USA 0.573
17 JAPAN 5.520 17 CANADA 0.639 17 SOUTH_KORE 0.578
18 TAIWAN 5.490 18 SLOVENIA 0.638 18 BELGIUM 0.583
19 SPAIN 5.312 19 SOUTH_KORE 0.636 19 SPAIN 0.625
20 SWITZERLAN 4.224 20 PORTUGAL 0.633 20 TAIWAN 0.627
As shown in Figure 1, there were several countries connected to the U.S that had the
highest degree and betweenness centrality values. Although the U.S. had some strong ties
with Asian countries (e.g., Singapore, South Korea, China, and Taiwan), it had a relatively
sparse network with European countries (e.g., Switzerland, Denmark, Spain, the Netherlands,
and France). Note that the thickness of a line indicates the frequency of co-authorship
relationships between two countries. The visualization is based on Kamada and Kawai’s
(1989) spring-embedding algorithm, in which the closer a pair of countries is in the two-
dimensional space, the stronger their relationship. The line length does not directly represent
the total number of co-authored papers, but because of the Kamada-Kawai algorithm, the
shorter the line, the higher the number of these papers.
Figure 1. International Co-Authorship Network of Big Data Research
15

RQ 2 - Semantic Network Analysis
Semantic Structure of Paper Titles
Before visualizing the semantic network, we calculated the frequency of specific
terms included in paper titles (for a detailed explanation of the semantic network procedure,
see Cho, Choi & Park, 2012). The 10 most frequently occurring terms included “data mining”
(76 times), “data” (56), “using” (54), “analysis” (29), “mining” (28), “application” (27),
“cloud computing” (24), “based” (23), “algorithm” (20), and “approach” (20). In addition, 11
terms occurred fewer than 20 times: “cyberinfrastructure” (13), “data integration” (12),
“MapReduce” (7), “data warehouse” (5), “computational science” (4), “big data” (2), and
“big science” (1). No title included the terms “Hadoop” and “NoSql.”
Figure 2 shows a semantic network using the top 50 terms and four terms (“data
warehouse,” “computational science,” “big data,” and “big science”) used for retrieving
papers related to big data. We organized the layout by using the principal-component option
in NetDraw (Borgatti, 2012). The thickness of a line indicates the strength of semantic
connections (in terms of co-occurrences) between two terms in a document. Here we
normalized this thickness by using the cosine of concurrence frequency. The results indicate
16

four clusters: “data collection,” “data storage,” “data processing,” and “data analysis and
visualization.”
Figure 2. Semantic Network of Paper Titles in Big Data (50 Most Frequently Occurring
Terms with the Cosine ≥ 0.1)
As shown in Figure 3, we used both the 50 most frequently occurring terms and the
top-20 countries in terms of degree centrality for the visualization of the semantic and
country network in a cosine-normalized network (Leydesdorff, 2010a and b). In the top left
corner, the U.S. stands alone by being connected to the most frequently used term—“data
mining”—and China is also visible by being connected to the second most frequently
occurring term— “data.” Noteworthy is that countries in Northern Europe, including Norway,
Denmark, and Sweden, formed a group with the U.K. and other European countries. The
prominence of the very specific term “gene” in this cluster reflects a general preference
among European researchers (e.g., bioinformatics) and the direction of national policies. In
17

the right corner, there is a large cluster in which several terms and countries are weakly
connected.
Figure 3. Semantic Network of Paper Titles and Countries in Big Data (50 Most Frequently
Occurring Terms and the Top 20 Countries with the Cosine ≥ 0.2)
Discusssion and Conclusions
The present study provides a snapshot of any structural changes in the academic community
during the transition from theory-based knowledge discovery to research practices driven by
big data. Several studies have indicated some changes in the community with respect to
research based on big data during the transition period from traditional computer science
18

fields to other engineering, social, and natural science fields. However, internationally co-
authored papers in the field of big data researchhave generally focused on primary
technologies (e.g., “data mining”) applying classical techniques to data sets or to
programming and related database issues. This suggests that SCI papers do not necessarily
focus on conceptually new methodologies for analyzing and synthesizing massive data sets.
However, the results suggest the emergence of some new subjects such as “MapReduce,”
implying that researchers are gradually focusing on methodologies that make use of new
digital and/or statistical tools to manage the deluge of data.
Although we invested in sharpening the search, one can still doubt whether we were
able to retrieve this new field using our search terms. Do the results meaningfully describe an
area of science? Emerging fields have both geographic and cognitive components. As shown
in the geographic map (Figure 1), the U.S. was central in various aspects because of its
connections with E.U. member countries as well as individual Asian countries. The network
shows various European countries with the second most central positions based on centrality
measures. In terms of structural hole indicators, however, some smaller and less advanced
countries were more “efficient” than “effective” in terms of controlling central positions. This
may be due to the mobility of doctoral candidates and postdoctoral fellows from these
countries to the U.S. or U.K. or perhaps more generally to Western Europe.
The semantic map (Figure 2) suggests the thematic structure of the major concepts of
the emerging field in terms of the (1) collection, (2) storage, (3) processing, and (4)
analysis/visualization of data. These principal components in the data matrix reveal the
cognitive structure of co-authorship at the global level. However, their coupling to the
geographic network (Figure 3) indicates an emerging division of labor in which “data” and
“data mining” are related to the U.S. and also to China to some extent. On the other hand, the
E.U. and Israel formed a separate cluster that was less specific in its vocabulary except for the
term “gene.” The rest of the world and other concepts tended to be blended into the third
group that was more heterogenous. The general lack of connections in the U.S. and E.U.
networks in terms of dominant semantics may reflect not only some differences in policy
incentives across the Atlantic but also a relatively weak intellectual framework in emerging
semantics. Note that central terms such as “big data” were part of the heterogeneous group.
The results suggest that a combination of words and locations in a two-mode network
can provide a richer representation of the emerging field of big data science than the sum of
19

two representations (Leydesdorff, 2010a, 2010b). This has long been known for combining
words as indicators of variations and citations as those of continuity and changes along time
axes (e.g., Braam et al., 1991; Lucio-Arias & Leydesdorff, 2009), but in the case of dispersed
fields or those fields driven by policy initiatives, a combination of geographic and semantic
perspectives may be more informative. Therefore, such a perspective can complement
informetric research in addressing questions concerning the emergence and interdisciplinarity
of big data research.
20

References
Abbasi, A., Chung, K.S. K., & Hossain, L. (2012). Egocentric Analysis of Co-
authorship Network Structure, Position and Performance. Information Processing &
Management, 48 (4), 671–679.
Anderson, C. (2008) The End of Theory: The Data Deluge Makes the Scientific
Method Obsolete. WIRED. 23 June, 2008.
http://www.wired.com/science/discoveries/magazine/16-07/pb_theory
Arzberger, P.,Schroeder, P., Beaulieu, A., Bowker, G., Casey, K., Laaksonen, L.,
Moorman, D., Uhlir, P., & Wouters, P. (2004a). An international framework to promote access
to data. Science. 303, pp. 1777–1778.
Arzberger, P.,Schroeder, P., Beaulieu, A., Bowker, G., Casey, K., Laaksonen, L.,
Moorman, D., Uhlir, P., & Wouters, P. (2004b). Promoting access to public research data for
scientific, economic, and social development. Data Science Journal. 29(3), pp. 135–152.
Bonacich, P. (2004). The Invasion of the Physicists. Social Networks 26(3): 285-288
Borgatti, S. P. (2012). Netdraw Network Visualization. Analytic Technologies:
Harvard, MA.
Borgatti, S. P., Everett, M. G. & Freeman, L. C. (2002).Ucinet for Windows: Software
for Social Network Analysis. Massachusetts: Analytic Technologies.
Braam, R. R., Moed, H. F., & van Raan, A. F. J. (1991). Mapping of science by
combined co-citation and word analysis. I. Structural aspects. Journal of the American
Society for Information Science, 42(4), 233-251.
Burt, R. S. (1992). Structural Holes: The Social Structure of Competition.
Cambridge, MA: Harvard University Press.
Cho, S. E., Choi. M. G., Park, H. W. (2012). Government-civic group conflicts and
communication strategies: A text analysis of TV debates on Korea’s Import of U.S. Beef.
Journal of Contemporary Eastern Asia, 11 (1). 1-20.
Freeman L. C., Borgatti, S. P., & White, D. R. (1991). “Centrality in valued graphs: A
measure of betweenness based on network flow.” Social Networks 13, 141-154.
Gilpress (April 26, 2012). A very short history of Data
Science.http://whatsthebigdata.com/2012/04/26/a-very-short-history-of-data-science/
21

Graham M. 2012. Big data and the end of theory?. The Guardian
http://www.guardian.co.uk/news/datablog/2012/mar/09/big-data-theory
Halevi, G., & Moed, H. F. (2012). The Evolution of Big Data as a Research and
Scientific Topic: Overview of the Literature. Research Trends. Issue 30. September.
http://www.researchtrends.com/issue-30-september-2012/the-evolution-of-big-data-as-a-
research-and-scientific-topic-overview-of-the-literature/.
Hanneman, R. A., & Riddle, M. (2005). Introduction to social network methods.
Retrieved February 20, 2006, from http://faculty.ucr.edu/~hanneman/.
Harris, R. (2012). International Council for Science (ICSU) and the challenges of big
data in science. Research Trends, 30, 11–12. http://www.researchtrends.com/issue-30-
september-2012/international-council-for-science-icsu-and-the-challenges-of-big-data-in-
science/
Hey, T., Tansley, S., & Tolle, K. (2009) The Fourth Paradigm. “Data-intensive
scientific discovery,” Microsoft.
King, G. (2011). Ensuring the Data-Rich Future of the Social Sciences. Science 11
February 2011: Vol. 331 no. 6018 pp. 719-721 DOI: 10.1126/science.1197872
Kwon, K.-S., Park, H.W., So, M.H., & Leydesdorff, L. (2012). Has Globalization
Strengthened South Korea’s National Research System? National and International Dynamics
of the Triple Helix of Scientific Co-authorship Relationships in South Korea.
Scientometrics.90 (1), 163-176.
Leydesdorff, L., Wagner, C., Park, H. W., & Adams, J. (2013). International
collaboration in science: The global map and the network. El professional de la información.
22 (1), 87-94.
Leydesdorff, L. (2007). ‘‘Betweenness centrality’’ as an indicator of the
‘‘interdisciplinarity’’ of scientific journals. Journal of the American Society for Information
Science and Technology, 58(9), 1303–1309.
Leydesdorff, L. (2010a). Eugene Garfield and Algorithmic Historiography: Co-
Words, Co-Authors, and Journal Names. Annals of Library and Information Studies, 57(3),
248-260.
Leydesdorff, L. (2010b). What Can Heterogeneity Add to the Scientometric Map?
Steps towards algorithmic historiography. In M. Akrich, Y. Barthe, F. Muniesa & P. Mustar
(Eds.), Débordements: Mélanges offerts à Michel Callon (pp. 283-289). Paris: École
22

Nationale Supérieure des Mines, Presses des Mines.
Loukides, M. (2012). What is data science? The future belongs to the companies and
people that turn data into products. Sebastopol, CA.: O’Reilly Media.
http://radar.oreilly.com/2010/06/what-is-data-science.html#data-scientists
Lucio-Arias, D., & Leydesdorff, L. (2009). The Dynamics of Exchanges and
References among Scientific Texts, and the Autopoiesis of Discursive Knowledge. Journal of
Informetrics, 3(2), 261-271.
McKinsey Global Institute (2011). Big data: The next frontier for innovation,
competition, and productivity. McKinsey & Company. www.mckinsey.com/mgi.
Miller, H.E. (2013). "Big-data in cloud computing: A taxonomy of risks" Information
Research, 18(1), paper 571. [Available at http://InformationR.net/ir/18-1/paper571.html]
Naur, P. (1974). Concise Survey of Computer Methods, Lund: Sweden:
Studentlitteratur AB.
OECD (2012). OECD Technology Foresight Forum 2012 - Harnessing data as a new
source of growth: Big data analytics and policies. OECD Headquarters, Paris, France
22 October 2012
O’Reilly Radar Team (2012). Planningfor Big Data: A CIO’s Handbook to the
Changing Data Landscape. Sebastopol, CA: O’Reilly Media.
Otte, E., & Rousseau, R. 2002. Social network analysis: A powerful strategy, also for
the information sciences. Journal of Information Science 28: 441–53.
Park, H. W. (2010). Mapping the e-science landscape in South Korea using the
webometrics method. Journal of Computer-Mediated Communication, 15, 211-229.
Park, H. W. (2012). How do social scientists use link data from search engines to
understand Internet-based political and electoral communication. Quality & Quantity. 46 (2),
679-693.
Park, H. W., & Leydesdorff, L. (2010). Longitudinal Trends in Networks of
University-Industry-Government Relations in South Korea: The Role of Programmatic
Incentives. Research Policy. 39 (5), 640-649.
Park, H. W., & Leydesdorff. L. (2008). Korean journals in the Science Citation Index:
What do they reveal about the intellectual structure of S&T in Korea? Scientometrics. 75 (3),
439-462.
23

Rogers, S. (2011). Data journalism at the Guardian: What is it and how do we do it?
http://www.guardian.co.uk/news/datablog/2011/jul/28/data-journalism
Rousseau, R. (2013, February 27). Keyword selection for big data papers. Personal
communication.
Rousseau, R. (2012). A view on big data and its relation to Informetrics. Chinese
Journal of Library and Information Science, 5 (3), 12-26. Savage, M., & Burrows, R. (2007).
The coming crisis of empirical sociology. Sociology, 41(5), 885–899.
Science staff (2011). Challenges and opportunities. An introduction to special issue.
Science, 331 (6018), 692-693. DOI: 10.1126/science.331.6018.692
http://www.sciencemag.org/content/331/6018/692.short
Shapiro, M., & Park, H. W. (2012). Regional Development in South Korea:
Accounting for Research Area in Centrality and Networks. Scientometrics. 90 (1), 271-287.
Snijders, C., Matzat, U., Reips, U.-D. (2012). ‘Big Data’: Big Gaps of Knowledge in
the Field of Internet Science. International Journal of Internet Science. 7 (1), 1–5.
Wagner, C. S. (2008). The New Invisible College. Washington, DC: Brookings Press.
Wagner, C. S., Leydesdorff, L., & Adams, J. (in preparation). Policy Implications of
the Global Network of Science.
24

Decomposing Social and Semantic Networks in Emerging “Big Data” Research

More Related Content

What's hot (20)

Similar to Decomposing Social and Semantic Networks in Emerging “Big Data” Research (20)

More from Han Woo PARK (20)

Recently uploaded (20)

Decomposing Social and Semantic Networks in Emerging “Big Data” Research