Network analysis of scientific collaboration in North Korea

Article information

Sci Ed. 2019;6(1):25-34

Publication date (electronic) : 2019 February 20

doi : https://doi.org/10.6087/kcse.152

Hyung Wook Choi^1,*, Ye Jin Choi^1,*, Soon Kim^,2

¹Department of Library and Information Science, Ewha Womans University, Seoul, Korea

²Research Institute for Social Science, Ewha Womans University, Seoul, Korea

Correspondence to Soon Kim soonkim123@ewhain.net

*These two authors contributed equally to this work.

Received 2019 January 22; Accepted 2019 February 8.

Abstract

Introduction

North Korea has maintained a very secretive status and remains isolated globally. Although North Korea has invested in scientific research, few selected research results have been published in international journals. However, North Korean leader Kim Jong-un announced that he intends to boost North Korea’s economy through science and education by having ‘a scientific and technical power and a talent power’ at a China visit in April 2018 [1]. With this movement, some scientists expect North Korea to open the door for more international research collaboration.

There have been mentions of North Korea’s position and research areas in specific academic fields. However, very few studies have examined research articles published by North Korean researchers with bibliometric analysis [2-4]. Bibliometric analysis is a research method that helps to clarify research trends and specific research areas within a particular academic field. However, bibliometric research about North Korea has some challenges due to the small number of published articles and the misclassification of various names for South Korea and North Korea [2]. As North Korea may be known as the Democratic People’s Republic of Korea, North Korea, DPRK, or DPR Korea, the author searched for “North Korea,” “DPR Korea,” and “North Korea” in the address field on the Web of Science [3]. Jeong and Huh [4]’s study result showed that Kim Il Sung University researchers had published the most articles and their main areas of research were physics, mathematics, materials science, chemistry, and engineering. China, Germany, and Australia were the main cooperating countries and the funding agencies were mainly Chinese. However, these studies primarily used quantitative and suggested statistical results.

In this paper, network analyses with author keywords and the co-authors of articles that were published by North Korean researchers are conducted. The limited number of published articles from North Korea means that co-word analysis and co-author analysis, which identify patterns in sub-areas with titles, abstracts, and keywords, would be a better methodology than citation analysis. From this result, visualized networks of core subject areas and primary authors of North Korea could be presented. This could bring a more in-depth view from learning more about science research in North Korea by analyzing the North Koreans’ scientific literature. Furthermore, this study result could provide the possibility of cooperation for those looking for opportunities to enter into research collaborations with North Korean researchers.

Methods

In this study, co-word analysis is carried out to identify the intellectual structure of studies from North Korea. All available data from Core Collection of Science Citation Index Expanded, Social Science Citation Index, and Arts & Humanities Citation Index in the Web of Science were collected and analyzed. Detailed information of published articles for 1976–2018 was collected through a country code search using the keyword “North Korea” (Dataset 1). After that, author keywords and the authors of each data were extracted to perform co-word analysis. To visualize the results, the structure of the median centrality of words and the parallel nearest neighbor clustering (PNNC) analysis of words were observed. The same process was conducted with author names.

Basic preprocessing and information extraction of data were performed with Bibexcel ver. 2014-03-25 (Persson O, available at: https://homepage.univie.ac.at/juan.gorraiz/bibexcel/index.html). In addition, COOC, which is a co-occurrence matrix generation program, and WNET (Lee JY, Seoul, Korea), which performs weighted network analysis, were used to obtain matrices for co-occurrence matrices, determine weighted network centrality through co-occurrence analysis using author keywords, and extract authors from the data. In addition, the network was visualized through NodeXL (Microsoft, Seattle, WA, USA) to understand its structure and scholarly communication.

First, to understand the detailed subjects and scholarly relations among authors in a field, a country code search was conducted with the keyword “North Korea.” As a result, 638 data was collected in total as of December 6, 2018. There is some misclassification of papers among South Korea and North Korea, so the authors’ affiliation addresses were checked manually.

Second, author keywords (DE: Author Keywords) and authors’ full names (AF: Author’s Full Name) were extracted using Bibexcel for co-word and co-author analysis. The collected keywords and names were capitalized with the first letter of each word. The number of keywords and names were defined with frequency of 3 and 5 respectively. Fifty-five keywords and 57 authors were selected as the final data for cooccurrence analysis as shown in Table 1.

Table 1.

Fifty-five final keywords with frequency >3

Third, in the case of authors with a family name, data were collected with the full registered names to prevent other names being misidentified as the same author when initialized. As a result, 2,156 author names in total were identified. Afterward, for the convenience of analysis, only the authors of frequency ≥5 were considered as the final analysis targets. Only those 59 authors were manually rechecked and one author whose full name and initials were confirmed was revised, finally confirming the final author list with 56 authors (Table 2).

Table 2.

Fifty-six final author names with frequency ≥5

In a co-occurrence matrix that applies the frequency, relationships do not appear between the key node and the non-key node [5]. Therefore, it is not suitable for network analysis when it needs to express the weight of the strength’s connection [6]. For network analysis in this paper, second-order Pearson’s correlation coefficient matrix (Pearson’s matrix) was used.

Pearson’s matrix can measure the similarity of co-occurrence patterns between two keywords and a third keyword [7]. The result of the Pearson’s correlation coefficient has a value between 1 and -1, where 1 indicates entirely related, 0 means that there is no association, and -1 means that they are completely inversely related. The higher the value, the higher the topic relevance between the two words and lower the value, the lower their connection.

Network analysis was performed to visualize the relationship between the keywords and authors to classify them into clusters according to similarity. For this, the Pathfinder network (PFNet) technique was applied to Pearson’s matrix and a network was constructed that left only essential links for each node. Afterward, PNNC was applied to subdivide the networks and the NodeXL program was used for visualization.

Results

Keyword network analysis

Analyzing the PNNC cluster of the co-occurrence word network using the Pearson’s matrix obtained from the 55 final keywords confirmed that the corresponding network was composed of 17 sub-clusters under four areas as shown in Table 3.

Table 3.

Parallel nearest neighbor clustering cluster grouping of keywords within the network

To identify the relationship between keywords, the PFNet was applied to Pearson’s matrix. Afterward, two types of centrality were measured to clarify which node is the main or core node within the network. First, the relative Triangle Betweenness Centrality (rTBC) is the centrality that measures a broad relationship by connecting other keywords and influential positions within the network. Second, relative Nearest Neighbor Centrality is the centrality of how much of an intermediary role it plays among other nodes within the network. The top 10 keywords for each centrality were compared to corroborate the core keywords (Table 4). Due to having the same ranks, 14 and nine keywords were analyzed for each centrality.

Table 4.

Centrality comparisons among top keywords within the network

For visualization, the rTBC of keywords was set to the size of the nodes. In addition, to express the PNNC cluster, the areas were set with the shape of the node and the clusters were to the color of the node (Fig. 1).

Fig. 1.

Network visualization based on parallel nearest neighbor clustering clusters with 55 keywords.

Author network analysis

The results of analyzing the PNNC cluster of the co-occurrence word network using the Pearson’s matrix obtained from 56 final authors confirm that the corresponding network is composed of 15 sub-clusters under four areas as shown in Table 5.

Table 5.

Parallel nearest neighbor clustering cluster grouping of authors within the network

The same analysis that was used for keyword network analysis was conducted to identify the relationship between keywords. After applying PFNet to Pearson’s matrix, rTBC and relative Nearest Neighbor Centrality were measured and the top 10 authors for each centrality were compared (Table 6).

Table 6.

Centrality comparisons among the top authors within the network

In addition, to clarify the distribution and relationships among the authors based on country and institution, author affiliation addresses were extracted (Figs. 2, 3).

Fig. 2.

The institution distribution of authors within the network.

Fig. 3.

The country distribution of authors within the network.

For visualization, the rTBC of the author was set to the size of the nodes. In addition, to express the PNNC cluster, areas were set with the shape of the node and the clusters were the color of the node (Fig. 4). Number of publication in Fig. 5.

Fig. 4.

Network visualization based on the parallel nearest neighbor clustering clusters with 57 authors.

Fig. 5.

Number of publication from North Korea searchable in Web of Science Core Collection by year.

Discussion

North Korea is one of the most closed-off countries in the 21st century, even with its recent interactions with other nations. Although limited research articles by North Korean researchers have been published and are available to the public, bibliometric analysis can be useful for getting an overview of the academic intellectual structure in North Korea.

Based on the tendency of increasing publications from North Korea as shown in Fig. 5, it is expected that the North Korean government will encourage researchers to publish research results in international journals [3]. North Korean researchers have expanded their publications to more than 50 articles since 2015.

Although the country’s research has focused on enhancing military strength, North Korean researchers have been publishing in other fields such as materials science, physics, and mathematics [8].

Based on keyword analysis results, the most researched academic topics were incline, compressible Navier-Stokes equations, quantum dot, switching, existence, Navier-Stokes equations, mechanical properties, Democratic People’s Republic of Korea, spark plasma sintering, Ent-Kaurane, keratin, DPRK, China, fractal interpolation function, metal nanocomposites, Fe2TiSi, and water treeing. The top three subject areas were mostly related to physics, chemistry, and mathematics.

The representative authors of sub-cluster were Jang Yong Man, Choe Song Hyok, Ho Kum Song, Choe Song Il, Sin Chung Sik, Kim Jongnam, Chang Xulu, Peng Fang, Choe Chol Ung, Duan Jingkuan, Choe Chunsik, Kronfeldt Heinz Detlef, Jin Hak Son, Ju Kyong Sik, and Sim Kyong Ho. Among these 15, eleven were from North Korea, three were from China, and one was from Germany. This shows that China most frequently conducts collaborative research with North Korea. All the eleven authors who have high betweenness centrality were included in C-9 sub clusters.

This study has some limitations: it lacks content analysis to clarify the specific relationships among subject areas and this study does not represent the research intellectual structure and author analysis within North Korea since this includes articles that were collaboratively created with foreign nations.

However, since very few studies have focused on North Korea’s research areas and authors, the results can lead to further research focusing on domain-oriented study to explore North Korea’s future research trends and changes.

Data Availability

Dataset 1. Original dataset for bibliometric scholarly network of North Korea is available from the Harvard Dataverse at: https://doi.org/10.7910/DVN/273J7G.

Article information Continued