2020 13th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI)

Data Analytics for Artificial Intelligence Research

from 2018 to 2020
Liying Zhou, X iaom in Li, Y i Liu, W enge Zuo
China Agricultrual University Library
Beijing, 100193, China

Abstract—This paper is based on literature dataset about Citation Reports (JCR). AI is one category in JCR categories
Artificial Intelligence from SCI and El. A series of indices, such and one of seven categories related to Computer Science.
as Documents, Times Cited, CNCI, Highly Cited Papers, Hot Engineering Index (El) is the broadest and most complete
Papers and El Controlled Terms are used to analyze the research engineering literature database available in the world. It
status and trends in the field of artificial intelligence in 2018-2020. provides a truly holistic and global view of peer reviewed and
Based on Documents, Times Cited and CNCI, high-yield indexed publications with over 20 million records from 77
countries, high-yield institutions, high-impact countries and high­ countries across 190 engineering disciplines. Every record is
impact institutions are identified. Based on Highly Cited Papers, carefully selected and indexed using El Thesaurus to ensure
Hot Papers and El Controlled Terms, the most productive topics
discovery and retrieval of engineering-specific literature that
and the most influential topics in AI subject are identified. The
results show that: AI is the third most productive sub-field in the engineering students and professionals can rely on. El
Computer Science, and it produces the most highly cited papers classification code is a numerical hierarchy of general subject
and hot papers; the three countries with most total paper output categories related to El Thesaurus. The classification code of
are China mainland, USA, and Japan, while the top three AI is “723.4”.
countries with highest average paper impact are USA, England B. Methodology
and United Kingdom; China mainland has the most high-yield
institutions, among which Tsinghua University ranks first; the This study uses SCI and El as data source, and retrieves
most influential topics discussed in highly cited papers are documents about AI published in 2018-2020. In Web of
Decision Making, Neural Networks, Convolution, Fuzzy Sets, Science, searching by “WC=(Artificial Intelligence) AND
Deep Learning, Learning Algorithms, etc. PY=(2018-2020)” can retrieve documents about AI published
Keywords-artificial intelligence; highly cited papers; E l in 2018-2020. In El, searching by “(723.4 WN CL) AND
Thesaurus; research topics (2018-2020 WN YR)” can retrieve relevant results about
“Artificial Intelligence” published in 2018-2020. Indicators
I. Introduction used in this paper were listed in Table I. Among them,
Artificial Intelligence (AI) is a technical science that studies Documents, Times Cited, Category Normalized Citation
and develops theories, methods, technologies and application Impact (CNCI), Highly Cited Papers, Hot Papers were defined
systems for simulating, extending and expanding human based on data set from SCI[3], while Controlled Terms came
intelligence[l]. The aims of AI is to imbue software with the from EI[4],
ability to analyze its environment using either predetermined
rules and search algorithms, or pattern recognizing machine TABLE I. Indicators used in this paper and their meaning
learning models, and then make decisions based on those Indicators M eaning
analyses. AI technologies are becoming useful as alternate Documents The volume o f three documents types, i.e. article,
approaches to conventional techniques or as components of review and proceedings paper.
integrated systems, and have been received increasing attention Times Cited The volume o f documents cited by other.
from academia and industry[2]. The research content o f AI CNCI It was calculated by dividing the actual count of
citing items by the expected citation rate for
includes knowledge representation, automatic reasoning and documents with the same document type, year of
search methods, machine learning and knowledge acquisition, publication and subject area.
knowledge processing systems, natural language understanding, Highly Cited The volume o f papers with top 1% times cited in
computer vision, intelligent robots, automatic programming, Paper each o f the 22 Essential Science Indicators (ESI)
and many other aspects. subject areas , per year.
Hot Papers The volume papers in the top 0.1% worldwide that
II. Data and M ethod were published in the last two years, based on
citation activity in the most recent two-month
A. Data source period, per ESI subject field.
Controlled Terms The vocabulary used to standardize the document
Web o f Science is a publisher-independent global literature ____________________ terms and index articles in El.____________________
database in the world, and it allows users to track ideas across As a unique identifier for the document, DOI was used to link data
disciplines and time from over 1.7 billion cited references from from SCI and El. The DOI of SCI highly cited papers about AI were
over 159 million records. Science Citation Index (SCI) in Web used to retrieve in El, and the El Controlled Terms of these highly
of Science is a citation database, which divides the documents cited papers were used to identify high impact research topics.
collected into more than 170 subject categories by Journal

This study is supported by the Fundamental Research Funds for the

Central Universities(SCUT) o f China under Grant 2019TC090.

978-0-7381-0545-1/20/$31.00 ©2020 IEEE 994

Authorized licensed use limited to: University of Gothenburg. Downloaded on December 20,2020 at 12:14:44 UTC from IEEE Xplore. Restrictions apply.
 III. Result and discussion B. Research Entities
A. Development Trends From 2018 to 2020, the top 10 countries/regions in AI are
listed in Table II. From the perspective of Documents, China
In SCI, Computer Science includes 7 sub-fields of Theory mainland, USA and Japan are in the top three. China mainland
& Methods, Information Systems, Artificial Intelligence, and USA belong to the first echelon, which has a very obvious
Interdisciplinary Applications, Software Engineering, advantage, and the number of papers is 2-3 times more than
Hardware & Architecture, Cybernetics. The development that of Japan. Japan is much closer to the following seven
trends of these 7 fields during 2018-2020 are shown in Figure 1. organizations, and the gap between them is not as great as that
The triangles represent the total output of papers, and the line of the first echelon. As for Times Cited, the total impact of
represents the highly cited papers. It can be seen from the papers, USA, China mainland and United Kingdom are in the
figure, the total output of papers in this field exceeded 300,000, top three. USA ranks first, far surpassing the second-ranked
with 1,856 highly cited papers, while the number of papers China by nearly twice. The gap between China mainland and
from AI field ranks third, close to 100,000; with the most United Kingdom is also very large, nearly double. Regarding
highly cited papers of 749. It can be concluded that papers in CNCI, the relative influence of each paper, USA, England and
AI have the greatest influence compared with other sub-fields. United Kingdom are in the top three. USA ranks first and the
gap between it and the following two is very significant.

TABLE II. Research performance of top 10 countries/ regions in

AI in 2018-2020

Country/Region Documents Times Cited CNCI

China mainland 201,644 1,236,647 0.97
USA 170,867 3,126,373 2.16
Japan 59,682 311,331 0.77
United Kingdom 54,607 772,445 1.56
Germany 46,058 494,643 1.51
England 45,757 672,336 1.64
India 42,376 228,592 0.79
France 42,297 459,291 1.28
Spain 35,797 320,589 0.97
Canada 32,590 449,181 1.43
Based on the signature information of the first authors or
Figure 1. Paper output in Computer Science and its seven sub-fields in 2018­
corresponding authors and the distribution of the authors’
2020 attribution, it can be seen that there are 172 high-yield
institutions with more than 100 papers in AI from 2018 to 2020.
The number of hot papers produced in the seven sub-fields Statistics on the regional distribution of high-yield institutions.
of Computer Science each year from 2018 to 2020 is shown in The top 9 countries/regions with high-yield institutions are
Figure 2. It can be seen from the figure that the total output of shown in Table III. It can be seen from Table III that China
hot papers in the AI field and the annual output are all listed on mainland has the largest number of high-yield institutions, with
the top. Moreover, the output of AI in 2018 and 2019 is nearly a total of 70. USA ranks second with 19 in total, and India
twice that of the second sub-field. The papers in AI had gained ranks third with eight. China mainland has become an absolute
much higher attention than other sub-fields. research force in AI.
2018 2019 2020 All
TABLE III. C ountry /Region distribution of high -yield
Artificial Intelligence 28 30 4 62 INSTITUTIONS OF AI IN 2018-2020
Interdisciplinary Applications 17 J6 1 34 Rank Country/Region N um ber of high-yield
Information Systems 6 18 4 28 Institutions
1 China mainland 70
Theoiy & Methods 7 15 2 24 2 USA 19
Software Engineering 3 6 1 10 3 India 8
4 Spain 7
Cybernetics 2 7 0 9 5 Hong Kong 5
Hardware & Architecture 2 4 1 7 6 South Korea 4
7 Portugal 4
8 Poland 4
Figure 2. Annual output of hot papers in Computer Science seven sub-fields 9 England 4
in 2018-2020
The top 10 institutions with the most documents or the
highest CNCI are shown in Table IV. From 2018 to 2020,
In 2018-2010, the most highly cited paper in AI is a study
on “semantic image segmentation with Deep Learning” among the high-yield institutions in AI, the top three
published in 2018, and which was cited 1,392 times cited in institutions of Documents are Tsinghua University in China
total[5]. It is also the most popular recent paper, with a mainland, Beihang University in China mainland, and
Shanghai Jiao Tong University in China mainland, with 721,
frequency of 204 times in the last 180 days.
547, and 539 papers respectively. The top three institutions of

995

Authorized licensed use limited to: University of Gothenburg. Downloaded on December 20,2020 at 12:14:44 UTC from IEEE Xplore. Restrictions apply.
CNCI are University o f Maryland College Park in USA, As can be seen from Figure 3, intelligent systems, Monte
Chinese University o f Hong Kong in Hong Kong, University Carlo methods, machine learning, reinforcement learning,
of Technology Sydney in Australia, with CNCI of 3.55, 3.46, neural networks, intelligent computing, learning systems, deep
and 3.43, respectively. Tsinghua University has the highest learning, learning algorithms and forecasting are the top ten
yield, University o f Electronic Science & Technology o f China research topics in AI from 2008 to 2020. The topics focused by
has the most influence, and the relative influence of each paper highly cited papers are decision making, neural networks,
of University of Maryland College Park is the highest. convolution, fuzzy sets, deep learning, learning algorithms,
machine learning, learning systems, numerical methods,
TABLE IV. Research performance of top 10 institutions in AI in genetic algorithms, etc. Although classification (of information)
2018-2020 and intelligence vehicle highway systems are two highly
Rank Institutions Documents Institutions CNCI productive topics in AI, there are no highly cited papers related
i Tsinghua University, 721 University of 3.55 to it.
China mainland Maryland College From 2018 to 2020, a total of 735 highly cited papers were
Park, USA produced in AI. The top institutions of highly cited papers are
2 Beihang University, 547 Chinese University of 3.46
China mainland Hong Kong, Hong listed in Table V. It can be seen that China mainland has the
Kong high-yielding institutions of highly cited papers. The top 11
3 Shanghai Jiao Tong 539 University of 3.43 institutions, except for the Thapar Institute o f Engineering &
University, China Technology Sydney, Technology in India and the University o f Granada in Spain,
mainland Australia almost all come from China mainland. University o f Electronic
4 Harbin Institute of 534 University o f Illinois 3.39
Technology, China Urbana-Champaign,
Science & Technology o f China tops the list with 40 highly
mainland USA cited papers, followed by 34 papers from Central South
5 Xidian University, 525 Southwest 3.36 University, and 26 papers from Northwestern Polytechnical
China mainland University —China, University.
China mainland
University of 494 Shandong University 3.17 TABLE V. TOP 11 INSTITUTIONS WITH HIGHLY CITED PAPER OUTPUT
Electronic Science & of Science &
Technology of Technology, China Rank Institutions Documents Country/Region
China, China mainland i University o f Electronic Science 40 China mainland
mainland & Technology o f China
Tianjin University, 479 University of 3.06 2 Central South University 34 China mainland
China mainland Electronic Science &
Technology of China, 3 Northwestern Polytechnical 26 China mainland
China mainland University
4 Thapar Institute of Engineering 23 India
Northwestern 477 Sichuan University, 3.00
& Technology
Polytechnical China mainland
5 Sichuan Normal University 23 China mainland
University, China
mainland 6 Sichuan University 21 China mainland
Huazhong University 470 Thapar Institute of 2.88
7 Harbin Institute of Technology 17 China mainland
of Science & Engineering &
Technology, China Technology, India 8 Wuhan University 15 China mainland
mainland
9 University o f Granada 14 Spain
Islamic Azad 450 Wuhan University, 2.79
University, Iran China mainland 10 Bohai University 14 China mainland
11 Liaoning University of 14 China mainland
C. Research Topics Technology
Based on El, the research topics o f AI in 2018-2020 are Based on highly cited papers, the high-impact research
identified, and the results are shown in Figure 3. The size of the topics of TOP11 institutions are analyzed, and a two-mode
font in the figure is proportional to the amount o f papers related network of the TOP institutions and their high-impact research
to the topic. The depth of the red font is proportional to the topics are created. In the two-mode network, the connection
amount o f highly cited papers related to the topic. The green threshold is set to 3, that means only when the institution has at
font indicates that the topic has only general related papers and least 3 highly cited papers on the topic will there be a
no high cited papers. connection between them. The two-mode network is shown in
Highly Cited Papers

" ne,it a"s“" .. Figure 4. The blue nodes in the figure are institutions, and the
D e e p L e a r n i n g VehicleH»0hwoy Systems M achine Learning size is proportional to the volume of highly cited papers of the
Classification (O f Inform ation) 1 • 1 ' •• corresponding institutions. The red nodes are high-impact
O p tim ira tio n research topics, and the size is proportional to the degree
“ ^■^Intelligent Sysl ems-..■ centrality o f the corresponding topic, that is, the number of
N e u r a l N e tw o r k s
Learning Systems,
Foatura Extraction ' • ’ institutions connected to the topic. The thickness of the line is
Learning Algorithms proportional to the volume o f highly cited papers of the
Decision M aking • • fo r e c a s t in g
Uncertainty Analysis Peep Neural Network» institution on the topic, and the thicker the line, the more highly
cited papers o f the organization on the topic.
Figure 3. Research topics in AI in 2018-2020

996

Authorized licensed use limited to: University of Gothenburg. Downloaded on December 20,2020 at 12:14:44 UTC from IEEE Xplore. Restrictions apply.
 paper “Some q-rung orthopair fuzzy Heronian mean
operators in multiple attribute decision making” [14]
are the most cited representative papers published by
Sichuan Normal University o f China on these two
topics.
• The dominant theme of Bohai University are adaptive
control systems. The paper “Adaptive Event-Triggered
Output Feedback Fuzzy Control for Nonlinear
Networked Systems With Packet Dropouts and
Actuator Failure”[15] is the most cited representative
paper published by Bohai University of China on this
topic.
• The dominant theme of University of Granada is
Figure 4. Two-mode network of the TOPI 1 institutions and their high­ decision making. The paper “Fuzzy Group Decision
impact research topics Making With Incomplete Information Guided by
Social Influence”[16] is the most cited representative
From the perspective of the organization layout, If the paper published by University of Granada of Spain on
topics of publishing more than 10 highly cited papers are this topic.
regarded as the dominant theme o f the institution, then we can • The dominant theme o f University of Liaoning
see from the two-mode network figure: University of Technology is adaptive control systems.
• The dominant theme o f University of Electronic The paper “Neural Networks-Based Adaptive Finite­
Science & Technology of China are decision making Time Fault-Tolerant Control for a Class of Strict-
and uncertainty analysis. The paper “Evidential Feedback Switched Nonlinear Systems’^ 17] is the
Supplier Selection Based on DEMATEL and Game most cited representative paper published by
Theory”[6] and the paper “Combining conflicting University o f Liaoning University of Technology of
evidence using the DEMATEL method”[7] are the China on this topic.
most cited representative papers published by For high-impact topics to institutions, the degree centrality
University of Electronic Science & Technology of of the topics like decision making, fuzzy sets, linguistics,
China on these two topics. mathematical operators, and numerical methods is relatively
• The dominant theme of Central South University are high, which means that multiple TOP institutions have
decision making and linguistics. The paper “A published at least 3 highly cited papers on them. The high­
projection-based TODIM method under multi-valued impact research and strong research background make these
neutrosophic environments and its application in topics have great development potential in AI.
personnel selection”[8] and the paper “Multi-criteria
decision-making method based on single-valued IV. Conclusion
neutrosophic linguistic Maclaurin symmetric mean This paper is based on from SCI o f Web of Science and El,
operators”[9] are the most cited representative papers combing with Documents, Times Cited, CNCI, Highly Cited
published by Centred South University o f China on Paper, Controlled Terms, to analyze research trend, research
these two topics. entities and research topics of AI subject. The analysis results
• The dominant theme of Sichuan University is decision show that the number o f highly cited papers and hot papers of
making. The paper “An approach to quality function AI are all greater than the other six sub-fields of Computer
deployment based on probabilistic linguistic term sets Science. Papers in AI have the greatest influence and attention.
and ORESTE method for multi-expert multi-criteria Although the volume of papers (Documents) in mainland
decision making”[10] is the most cited representative China and USA is relatively close, in terms o f influence, USA
paper published by Sichuan University of China has opened a large gap with China both in terms of total
mainland on this topic. influence (Times Cited) and average influence (CNCI). China
• The dominant theme of Thapar Institute of Engineering has the most high-yield institutions (more than 100 articles
& Technology are decision making, fuzzy sets and have been published by the first author or corresponding
mathematical operators. The paper “Generalized and author), and far exceeds USA. The top three institutions with
group-based generalized intuitionistic fuzzy soft sets most papers or most cited times are all from China Mainland,
with applications in decision-making”[ ll] and the while only one of the top three institutions with highest CNCI
paper “Linguistic Pythagorean fuzzy sets and its is from China Mainland. In 2018-2020, intelligent systems,
applications in multiattribute decision-making Monte Carlo methods, machine learning, reinforcement
process”[12] are the most cited representative papers learning, neural networks, intelligent computing, learning
published by Thapar Institute o f Engineering & systems, deep learning, learning algorithms and forecasting are
Technology of India on these three topics. the top ten research topics, while decision making, neural
• The dominant theme of Sichuan Normal University are networks, convolution, fuzzy sets, deep learning, learning
decision making and mathematical operators. The algorithms, machine learning, learning systems, numerical
paper “Pythagorean fuzzy power aggregation operators methods, genetic algorithms are the most influence topics.
in multiple attribute decision making”[13] and the Topics like decision making, frizzy sets, linguistics,

997

Authorized licensed use limited to: University of Gothenburg. Downloaded on December 20,2020 at 12:14:44 UTC from IEEE Xplore. Restrictions apply.
mathematical operators, and numerical methods are the most [8] Ji, Pu; Zhang, Hong-yu; Wang, Jian-qiang. A projection-based TODIM
influential high-impact research topics for the TOP institutions method under multi-valued neutrosophic environments and its
application in personnel selection. Neural Computing & Applications,
in AI. 2018, 29(1): 221-234.
[9] Wang, Jian-qiang; Yang, Yu; Li, Lin. Multi-criteria decision-making
A cknow ledgm ent method based on single-valued neutrosophic linguistic Maclaurin
The authors are grateftd for insightful discussion with symmetric mean operators. Neural Computing & Applications. 2018,
30(5): 1529-1547.
Professor Changshou Luo from Institute o f Agricultural [10] Wu, Xingli; Liao, Huchang. An approach to quality function deployment
Information and Economics, Beijing Academy of Agriculture based on probabilistic linguistic term sets and ORESTE method for
and Forestry. Wenge Zuo is the corresponding author of this multi-expert multi-criteria decision making. Information Fusion, 2018,
paper. 43: 13-26.
[11] Garg, Harish; Arora, Rishu. Generalized and group-based generalized
References intuitionistic fuzzy soft sets with applications in decision-making.
Applied Intelligence, 2018,48(2): 343-356.
[1] Hamet, Pavel; Tremblay, Johanne. Artificial intelligence in medicine. [12] Garg, Harish. Linguistic Pythagorean fuzzy sets and its applications in
Metabolism: clinical and experimental, 2017,69(S):S36-S40. multiattribute decision-making process. International Journal of
[2] Liu, Ruonan; Yang, Boyuan; Zio, Enrico; Chen, Xuefeng. Artificial Intelligent Systems, 2018, 33(6): 1234-1263.
intelligence for fault diagnosis o f rotating machinery: A review. [13] Wei, Guiwu, Lu, Mao. Pythagorean fuzzy power aggregation operators
Mechanical Systems and Signal Processing, 2018, 108: 33-47. in multiple attribute decision making. International Journal o f Intelligent
[3] Analytics Clarivate. Indicators Handbook[EB/OL]. [2020-05-16]. Systems, 2018, 33(1): 169-186.
https://incites.help.clarivate.com/Content/Indicators-Handbook/ih- [14] Wei, Guiwu, Gao, Hui, Wei, YU. Some q-rung orthopair fuzzy Heronian
about.htm. mean operators in multiple attribute decision making. International
[4] Elsevier. EI[EB/OL]. [2020-05-16]. Journal o f Intelligent Systems, 2018, 33(7): 1426-1458.
https://www.elsevier.com/solutions/engineering- [15] Zhang, Zhenxing; Liang, Hongjing; Wu, Chengwei; Ahn, Choon Ki.
village/content/compendex. Adaptive Event-Triggered Output Feedback Fuzzy Control for
[5] Chen, Liang-Chieh; Papandreou, George; Kokkinos, Iasonas; Murphy, Nonlinear Networked Systems With Packet Dropouts and Actuator
Kevin; Yuille, Alan L. DeepLab: Semantic Image Segmentation with Failure. IEEE Transactions on Fuzzy Systems, 2019,27(9): 1793-1806.
Deep Convolutional Nets, Atrous Convolution, and Fully Connected [16] Capuano, Nicola; Chiclana, Francisco; Fujita, Hamido; Herrera-Viedma,
CRFs. IEEE Transactions on Pattern Analysis and Machine Intelligence, Enrique); Loia, Vincenzo. Fuzzy Group Decision Making With
2018, 40(4): 834-848. Incomplete Information Guided by Social Influence. IEEE Transactions
[6] Liu, Tianyu; Deng, Yong; Chan, Felix. Evidential Supplier Selection on Fuzzy Systems, 2018, 26(3): 1704-1718.
Based on DEMATEL and Game Theory. International Journal o f Fuzzy [17] Liu, Lei; Liu, Yan-Jun; Tong, Shaocheng. Neural Networks-Based
Systems. 2018,20(4): 1321-1333. Adaptive Finite-Time Fault-Tolerant Control for a Class of Strict-
[7] Zhang, Weiquan; Deng, Yong. Combining conflicting evidence using Feedback Switched Nonlinear Systems. IEEE Transactions on
the DEMATEL method. Soft Computing, 2019, (23)17: 8207-8216. Cybernetics, 2019,49(7): 2536-2545.

998

Authorized licensed use limited to: University of Gothenburg. Downloaded on December 20,2020 at 12:14:44 UTC from IEEE Xplore. Restrictions apply.