=Paper=
{{Paper
|id=Vol-2318/paper23
|storemode=property
|title=Detection of Expert Groups for Scientific Expertise
|pdfUrl=https://ceur-ws.org/Vol-2318/paper23.pdf
|volume=Vol-2318
|authors=Iryna Balagura,Valentyna Andrushchenko,Ivan Gorbov
|dblpUrl=https://dblp.org/rec/conf/its2/BalaguraAG18
}}
==Detection of Expert Groups for Scientific Expertise==
https://ceur-ws.org/Vol-2318/paper23.pdf
Detection of Expert Groups for Scientific Expertise
Iryna Balagura1[0000-0001-9627-2091], Valentyna Andrushchenko2
and Ivan Gorbov1[0000-0001-6888-0866]
1
Institute for Information Recording of National Academy of Sciences of Ukraine, Kyiv,
Ukraine
2
National Information Center for Ukraine-EU S&T cooperation, Kyiv, Ukraine
balaguraira@gmail.com
Abstract. Today there is no unified model for selecting experts or specialists on
formal grounds, therefore this project proposes an approach to solving the prob-
lem of searching for experts based on networks of co-authors. The method of
research of scientific collectives is proposed, which allows to find groups of
scientists whose activities correspond to the topics on the basis of the network
of terms according to the given terms of a certain domain. The work of the al-
gorithm is realized on the basis of information from the Web of science, a com-
parison of the work of the algorithm with the system Aminer. The proposed me-
thod allows searching for more flexible requests. It is advisable to use the au-
thors' databases while searching for scientific groups to exclude identical names
and obtain detailed information.
Keywords: Co-author Networks, Co-word Networks, Experts, Scientific
Teams, Scientific Database.
1 Introduction
There is no unified model for formal searching of experts or specialists at present
time. Therefore, we propose the method of searching of experts using co-authors net-
works. The actual task is the selection of competent experts for the involvement of
scientists in solving of important public problems. Additionally, Ukrainian Govern-
ment passed the Law of Ukraine "On Scientific and Scientific and Technical Activity"
of December 26, 2015, which initiated changes in science in Ukraine, so the qualita-
tive expertise is also needed for developing of the science [1].The urgent for search
and formation of the expert groups are also associated with the Ukraine participation
in the program "Horizon 2020" and taking the place of Ukrainian scientists as experts
in competitive projects [2]. The defining of actual research directions where Ukraine
could represent main scientific results to the world community is one of the state
priorities. Formation of expert teams is necessary for this task. The proposed method
can act as an essential tool for objective decision-making in the creation of such
groups. Using the proposed method could help at objective search of international
experts to assess challenges and reform of individual sectors in the state based on the
272
respective scientific papers. Fast and skilled searching of qualified expert groups will
assist to the effective implementation of the Law of Ukraine "On scientific and tech-
nical expertise" and the Law of Ukraine "On innovation activity." [3].
Another actual task is searching experts for scientific collaborations. On the one
hand there are multiple ways to find the partners for forming the multidisciplinary
collaboration, and on the other hand, all these instruments are boiled down to genera-
lized ways to find the collaborators via the theme of research and short information on
the desired project and describing the work opportunities in research institutions.
There are such opportunities like Enterprise Europe Network cooperation opportuni-
ties database
(http://een.ec.europa.eu/tools/services/SearchCenter/Search/ProfileSimpleSearch).
CORDIS partner search platform (https://cordis.europa.eu/home_en.html). Partici-
pants Portal Partner Search (https://ec.europa.eu/research/ partici-
pants/portal/desktop/en/organisations/partner_search.html).
Enterprise Europe Network, cooperation opportunities database, provides an op-
portunity for the search by using the search line and precise options. CORDIS partner
search platform implies the usage of the text phrases to find the needed information
on the results of supported projects. The new instrument of the Horizon – 2020 partic-
ipants’ portal gives new opportunities for the grantees. It presumes to fill in several
options, including the keywords. There are other ways such as using a social network
for researchers (ORCID and Research Gate) and through the groups and proposals of
business- and employment-oriented service like LinkedIn.
However, there can be applied other attempts, which can allow more precise ap-
proaches to find correlations with current research and to widen the frames of the
work and its enlargement. Scientometrics – the measures of the research work through
investigating the publication activity of the author, research team or research institu-
tion. The information on publications can represent the basis for acquiring new arrays
of information and provide its further procession to receive new data which can ap-
pear new instruments for decision making. Citation indicators are used for describing
the nature and main statistics of research. Co-author networks are showed individuals
cooperation and could be used for predict next work.
The main idea of the study is to develop the method of qualified and formal detec-
tion of expert groups for scientific and technical expertise according to theirs skills
and publications.
2 Existing approaches to the search for expert teams
Expert searches can be found on many Ukrainian and foreign scientific, cultural and
other websites. It is a method of questioning, and it is conducted in an online or off-
line form [4]. For candidates, there are specific requirements for experience, skills and
job positions for examining one or another field. This expert search method is current-
ly the most widespread in the world. However, for not all types of work, it is possible
to announce a long open competition of tasks, and not in each case, the specialist will
273
pay attention to the announcement. Therefore, the search for experts through ques-
tionnaires and contests is not the most effective.
The expert searches are also possible using scientific databases, which contains
scientists profiles with information about the publications, citations and other scien-
tometric data. Scientific profiles can be found in Google Scholar, Scopus, Web of
science and other databases. There are also resources to combine information about
scientists from different databases, including ORCID (https://orcid.org/), "Bibliome-
trics of Ukrainian Science" (http://www.nbuviap.gov.ua/bpnu/ ), "Scientists of
Ukraine" (http://irbis-nbuv.gov.ua), Aminer (https://aminer.org/) and others. ORCID
is a worldwide service that provides a union of publications and identifications of
persons of the author. "Bibliometry of Ukrainian-science", "Scientists of Ukraine"
created by the VI Vernadsky National Library of Ukraine in the first case on the base
of scientific profiles from world scientific and technical resources, in the second case
- information on abstracts of papers. AMiner is a database of scientists from artificial
intelligence containing 130614292 scholars created in China. AMiner provides de-
tailed information about the author: distribution of publications according to topics,
citation, the h-index, g-index, collaboration network, place of work, key skills and
ranking among others in the scientific areas. Aminer conducts the sorting, for exam-
ple, the country, the language of publications, gender, authorship resources, h-indexes
and detailed sections, as well as a defined concept [6]. The main disadvantage of such
sites is the search restriction according to the identified thematic sections and the
absence of a full picture of scientific directions and teams.
In the paper [7] a comparative analysis of the methods of the automatic search of
experts was conducted. The expert search models are divided into probabilistic (com-
putations for authors or topics), ranking and network (PageRank, HITS). The choice
of method depends on the purpose of the expert search and data source. It is the input
data that most determine the result, so the first stage of choosing a database of publi-
cations is the most important.
3 The search method of expert teams
It is reasonable for seeking of experts for solving governmental tasks to select out of
the research teams that were formed within certain scientific schools and choose
specialists from different groups. The sources of experts could are big scientific
databases containing information on scientific activities in Ukraine. We propose to
use the analysis of co-authors networks for the definition of the scientific groups
using algorithms based on modularity [8].
The research deals with the investigation of co-authors and co-words networks
appliance for scientometric analysis of abstract databases for describing of the main
scientific areas structure. We propose to use the methods of co-author and co-word
network analyzes on the base of abstract databases. Co-author network is a network
structure where nodes are scientists and links are co-authorships, size of nodes and
width of lines are depends of network characteristics and common paper numbers.
274
Co-word networks could be built on occurred pairs of terms and shows their intercon-
nections. According to the algorithm from paper [9] terms will be extracted using
frequency characteristic in abstracts. Co-word and co-author networks could be used
for identification and description of scientific groups and research topics, the most
communicative researchers and main principles of science communication. For the
analysis we have to use main principles and instruments of complex networks, that
are decrypted in many works [9-16].
Co-word networks research teams publications allow to find common "narrowed"
line of research with a clearly defined system of concepts (terms); common terminol-
ogy may differ in detail from the general in a separate sciences; reduce the noise in-
formation that facilitates the work of experts in the knowledge that forms the model
domain.
We define the co-authorship network G = (V, E) as an object given by a pair of
sets (V, E), where V is the set of nodes (authors), E × V × V is the set of edges (co-
authorship links) . Then the network of terms can be represented in the form T = (C,
L), where C is the set of terms written by the authors of V, and L is the set of relation-
ships between terms. By connecting GUT we get a heterogeneous network H contain-
ing vertices V and C, the links E, L and M - the connection between the vertices V
and C (the relation the author used the terms in the publications). It is proposed to
explore the network of collaborators gradually and to select scientific groups and
leaders of scientific directions according to the subject, followed by a network of
terms to determine the map of the subject area. The combination of networks of co-
authors and network of terms makes it possible to identify scientific groups that are
most precisely included in the given problem.
On the first stage there are defined the field and the scientific concepts, by which
analyses are provided and then the review file is downloaded and filtered. The result
of the first and second stage is data about authors filtered by the several descriptors
and connections between them, i.e. the matrix appropriate of the network.
The next stage is forming of the co-authorship network according to the chosen
subject field, and the main characteristics and also the calculation of additional para-
meters. As the result, the main characteristics of scholar’s cooperation, scientific col-
laborations and the most communicative researchers in frames of the definite scientif-
ic concepts are defined.
Networks are divided into the clusters using the modularity measure. The mod-
ularity of a node is a value that evaluates the density of bonds in a coherent compo-
nent in comparison with the bundles between the components. In general terms, mod-
ularity can be defined as:
N
Q eii ai
i 1
where eij – the element of the matrix adjacency graph, equal to the ratio of the number
of edges, which combine two societies i and j, to the total number of edges in the
network, ai – the ratio of the number of edges connecting the vertices in the communi-
ty, to the total number of edges:
275
N
ai eij
j 1
The high modularity of the network indicates a strong connection in the clusters and
the weak link of the network itself [15].
The fourth stage is dedicated to the selection of full text publications of the most
communicative scholars and organizing of the text package to extract core terms
(words and word combinations) per scientific concepts. The visualization of terms
networks and the concept integrally, calculation of the main parameters are per-
formed. And generalization of results, description of the core characteristics, field
trends are also provided on this stage. Then we have to add two networks and form
heterogeneous network which consists of co-word and co-authored networks. The
father analysis of the union network could identify teams with the most relevant to the
concept researches and forecast possible cooperation. Also we propose to rank scien-
tists with centrality measures that allow detect the most collaborative scientists. The
identification of important nodes in the network is an urgent task and requires a de-
tailed study of the subject of research, since there are many coefficients that provide
versatile characterization of the vertices, and the feasibility of their application is
determined only with the purpose of the experiments. The degree centrality, which in
fact is an indicator of the number of articles in collaboration, reflects the volume of
work of the author, and the number of author ties characterize the circle of his co-
authors. In fig. 1 level of centrality of nodes corresponds to their value. Authors with
a high level of centrality index are linked in separate groups - scientific groups and
not related.
The degree centrality estimate the authors by their communicability and can be
used to predict the authors productivity. Centrality degree in weighted graph is:
C D (i ) ki(1 ) si ,
where ki – the sum of edges of between nodes:
N
ki mij
j 1
si – the sum of weights of the edges, α - coefficient which choosing depends on the
case [16]:
N
si ij
j 1
Betweenness centrality quantifies the number of times a node acts as a bridge
along the shortest path between two other nodes:
276
CB (i ) g jk (i )
jk
, i j, k.
where gjk(i) – number of shortest paths that crossing vertex i [16].
In the sense of scientific collaboration betweenness centrality provides the ability
to identify authors which are linking scholarly teams.
So, we propose search method of experts, which contains the following steps:
1. The definition of the field in which the search and related concepts are conducted.
2. The filtration of data from a scientific database.
3. The formation of a co-author network, the main characteristics of the network are
calculated, and the network is split into groups based on modularity.
4. The formation of a network of terms for each team of collaborators.
5. The formation of separate heterogeneous networks, which combine co-authors and
terms according to the clusters formed in 3.
6. The identification of teams with the most relevant concepts.
7. A ranking of scholars within individual teams is carried out according to the cen-
trality measures.
8. The expert list is formed accordingly.
4 Experimental Results and Discussion
We choose the Web of science database as a source of data. Web of science (Clarivate
analytics)consists of 8,700 carefully selected journals (Core Collection) and the cita-
tion rates of scientific literature. The main citation indicator of this platform is the
Impact Factor (Index of Influence) of the scientific publication. The Web of Science
(WoS) includes Science Citation Index Expanded (from the natural sciences), Social
Sciences Citation Index (from Social Sciences), Arts and Humanities Citation Index
(on Arts and Humanities), Emerging Index (new editions) and other. Science Citation
Index Expanded covers 6650 journals of 150 disciplines (astronomy, chemistry, biol-
ogy, biochemistry, mathematics, physics, medicine, science of materials, pharmacol-
ogy, etc.). Search depth since 1975. Social Sciences Citation Index - covers 1950
journals of 50 disciplines (anthropology, history, jurisprudence, linguistics, philoso-
phy, politics, psychology, sociology, etc.). Search depth since 1975. Arts and Human-
ities Citation Index - A Bachelor of Arts and Humanities. It covers 1160 journals in
the fields of art, folklore, history, linguistics, archeology, literature, music, philoso-
phy, poetry, religion, theater, radio and television. Search depth since 1975. The web
of science contains resecherID, which allows you to identify the author, to identify the
main scientific metrics and to rank among experts by citing, collaborating, and pub-
lishing activity.
The Web of Science was used as a database for the implementation of the algo-
rithm and the team of the National aviation university as a sample. In the abstract
database Web of Science 631 publications of the authors of the NAU were found. The
largest number of publications is devoted to aviation and technical sciences, in the
humanities and the arts the university is represented by two publications, which al-
277
most did not appear on the network of terms. The co-authors and a co-word networks
was created using the software VOSviewer [6], as shown in fig. 1. The nodes
represent the authors, and the edges show the co-authorship of these authors in ar-
ticles. The larger the node, the more articles published by the author. The network of
scientific cooperation assigns nodes to several clusters that describe scientific
groups [7]. The clusters are marked with different colors.
The relationship between the terms in publications is shown in Fig. 2, where each
node represents a term. The size of the node corresponds to the frequency with which
this term was used. Depending on the clusters by a frequency of simultaneous use the
terms are displayed in different colors. The same color is indicated by terms that are
often used together in one publication. The three most common terms of the network
are signals processing, signal radar, aircraft and others. The combination of co-author
networks and co-word networks makes it possible to identify the teams that work the
most on a given topic. At the next stage, it is possible to rank and allocate leaders of
scientific directions. The databases of scientists are used for person identification after
receipt on the basis of co-authorship and terms. The databases give an opportunity to
test the experience of authors, to separate authors with similar names
Fig.1.Co-author network using Web of science papers of the university team
The network was divided into groups based on the calculation of the modularity of
nodes. Modularity of a node is a value that evaluates the density of bonds in a cohe-
rent component in comparison with the bundles between the components. The pres-
ence of scientific groups that may have signs of scientific schools in the university
can be traced clearly to the feature of the majority of the click, which in its essence
contains a powerful author with a large number of articles in co-authorship and a
significant number of small nodes - the authors-students. In fig. 1 shows a fragment of
a network of collaborators' networks, which contains data on author Yanovsky, his
co-authors and co-authors of his co-authors.
278
Fig. 2. Co-word network using Web of science papers of the university team
For comparing of searching experts method, we use the concept link prediction in
the database Aminer. The Aminer database has found 66 profiles for the concept. In
each author's profile there is a percentage of papers of specific topics, but there is no
possibility to show the presence of individual concepts.
5 Conclusions
The method for detection of scientific teams is proposed, which allows finding groups
of scientists whose activities correspond to the topics by the network of terms accord-
ing to the given terms of a certain domain. It is shown that the proposed method al-
lows searching for more flexible requests in comparison with existing expert databas-
es. It is advisable to use the databases of scientists while searching for scientific
groups to exclude identical names and obtain detailed information about the expert.
The possibility the method usage was shown with examples from Web of science
database. The better results could be obtained with adding several international data-
bases.
The aim of the algorithm is to correct researches in Ukraine for activation of inter-
national research and popularity of papers which is depend of the topics. Using of the
algorithm could help scientists to get more authority in the world.
279
The method is a part of complex estimation of interdisciplinary and searching
priorities and scientific cooperation.
The future research will consist of studying of co-author networks according to
concepts with high interdisciplinary degree and developing of methods for searching
cooperation partners with data from Web of science, Scopus, Google Scholar. Ana-
lyzing of scientific databases allow to detect existing researches and to forecast the
possibilities of next cooperation of apart scientists or scientific teams.
The publication contains the results of research conducted under the grant of the
President of Ukraine under the competitive project F75 / 173-2018 of the State Fund
for Fundamental Research.
References
1. Ukrainian laws Homepage, http://nucpi.nas.gov.ua/news/item/67-zakon-ukraini-pro-
vischu-osvitu.html
2. Horizon 2020 Homepage, https://h2020.com.ua/
3. Ukrainian laws Homepage, http://zakon.rada.gov.ua/laws/show/51/95-%D0%B2%D1%80
4. Udding M.N., Duong T.H., Oh K.J., Jung J.-G., Jo G.-S.: Experts search and rank with so-
cial network: an ontology –based approach, International Journal of Software Engineering
and Knowledge EngineeringVol. 23, No. 01, pp. 31-50 (2013)
5. C. Wei, W. Lin, H. Chen, W. An, and W. Yeh. Finding experts in online forums for en-
hancing knowledge sharing and accessibility, Computers in Human Behavior, vol.51,
pp.325-335 (2015)
6. Robin Brochier,AdrienGuille, Benjamin Rothan,
JulienVelciImpactoftheQuerySetontheEvaluationofExpertFindingSystems, CoRR,
abs/1806.10813 (2018)
7. Lin, S., Hong, W., Wang, D., & Li, T. A survey on expert finding techniques. Journal of
Intelligent Information Systems, 49(2), 255–279.(2017)
8. Lande D.V., Andrushchenko V.B., Balagura
I.V.:FormationoftheSubjectAreaontheBaseofWikipediaService,OpenSemantic
Technologies forIntelligentSystems : conference, Minsk. – pp. 211-214. (2017 )
9. Dmitry Lande, AndreySnarskii, ElenaYagunovaThe Use Of Horizontal Visibility Graphs
To Identify The Words That Define The Information Structure Of The TextCEUR Work-
shop Proceedings. Vol-1108 urn:nbn:de:0074-1108-1. ISSN 1613-0073. Selected Papers of
the 15th All-Russian Scientific Conference "Digital libraries: Advanced Methods and
Technologies, Digital Collections" Yaroslavl, Russia, October 14-17, 2013. - P. 158-
164.(2013)
10. I.V. Gorbov, S.V. Kadenko, I.V. Balagura, D.Yu. Manko, O.V. Andriichuk. Elicitation of
possible scientific expert groups at co-authorship network using decision support methods
// Data Recording, Storage & Processing , Vol. 15, No 4, P. 77 – 85 (2013).
11. I. V. Balagura, D. V. Lande, I. V. Gorbov. Studying of node importance characteristics in
co-author network // Data Recording, Storage & Processing , Vol. 15, No 1, P. 45 – 52
(2013).
12. Lande D., Andrushchenko V., Balagura I.: Data Science in Open-Access Research On-line
Resources // Proceedings of the 2018 IEEE Second International Conference on Data
Stream Mining & Processing (DSMP) - pp. 17-20. Web of Science, Scopus (2018)
280
13. Lande D.V., Balagura I.V., Andrushchenko V.,B.: The detection of actual research topic
using co-word networks, Open Semantic Technologies for Intelligent Systems (OSTIS -
2018): conference, Minsk, - pp. 207-210 (2018)
14. Lande D.V. , Balagura I.V., Dubchak N.A.
TheDetectionofActualResearchTopicsinPhysicsforUkrainianScientificGroups // 2017
IEEE InternationalYoungScientistsForumonAppliedPhysicsandEngineering YSF-2017,
October 17-20, Lviv, P.64-66 (2017)
15. Zhao S. X., Rousseau R., Ye F.Y. H-Degree as a basic measure in weighted networks,
Journal of Informetrics. – V.5, № 4. – P.668–677.(2011)
16. Newman M. E. J. Coauthorship networks and patterns of scientific,Proceedings of the Na-
tional Academy of Sciences of the United States of America. – V.101,№1. – P.5200-5205.
(2004)