Who is the best connected researcher? An Analysis of Co-authorship Networks of Knowledge Management from 2000 to 2010 Mei-yun Zuo 1, Xiao-qing Hua1, Xiao-wei Wen1 1 Department of Economic Information Management, School of Information, Renmin University of China, Beijing, China (zuomy@ruc.edu.cn, vivian.hua@ruc.edu.cn, elowise0516@ruc.edu.cn) Abstract - Who is the best connected researcher? This study uses social network analysis to address this question. Specifically, 1831 papers published in six top knowledge management research journals from 2000 to 2010 are analyzed. Firstly, by analyzing the degree distribution of the network, we find there exist a few researchers who have quite a large cooperation network, and in contrast a large sum of common researchers who have just a few cooperators. Next, we respectively identify 20 authors who have the most connections with others according to degree centrality, who are in the core positions according to closeness centrality and who play the “bridge” roles in the network according to betweenness centrality. Finally, the biggest connected sub-graphs, the 57-Actors Graph is extracted and discussed. Keywords - Knowledge management, co-authorship network, social network analysis, individual status, academic groups I. INTRODUCTION Prior research demonstrates that there are many successful and prolific individuals producing novel, interesting, and high-quality knowledge management (KM) works in refereed journals [1] [2] [3] [4] or conference proceedings. With contributions of these productive individuals, the field of KM developed a strong foundation from which it has now flourished. Then we may wonder about who is the best connected researcher in the field of KM? Who are the most connected researchers in the field of KM? networks have been analyzed in other fields, such as library science [6], information retrieval [7], e-markets [8], medical [9] and so on, but few have studied co-authorship networks in KM discipline. In this study, based on analysis of KM’s co-authorship network’s structure, our objective is to identify various types of connected researchers and investigate collaboration patterns in different research groups extracted from the network. III. METHODOLOGY Social network analysis (SNA) is based on the premise that the relationships among social actors can be described by a graph. The graph’s nodes represent social actors and the graph’s edges connect pairs of nodes and thus represent social interactions. Given that we have established a social network graph, we can describe its properties on two levels, namely by global graph metrics and individual actor properties. Global graph metrics seek to describe the characteristic of a social network as a whole, for example the degree distribution, graph’s average path length (APL), clustering coefficient, the number of connected sub-graphs, etc. Actor properties relate to the analysis of the individual properties of network actors, e.g. actor status in a cluster. The status of an actor is usually expressed in terms of its centrality, i.e. a measure of how central the actor is to the network graph. In this article, we will combine the use of two SNS software tools Ucinet and Pajek to perform the analysis of co-authorship network. IV. DATA COLLECTION AND PRE-PROCESS II. LITERATURE REVIEW Recently, a number of researchers have realized the importance and necessities to understand the identity of KM’s research community and concentrate their focuses on it. Serenko and Bontis identified a number of very productive individuals and institution on KM/IC (intellectual capital) [5], developed a global ranking of KM/IC academic journals [2], and investigated the presence of the superstar effect in the KM/IC academic discipline [4]. Although the scientific collaboration is also a prominent indicator of KM’s research community, few papers on scientific collaboration in KM had been published. Lately with the development of social network analysis methods, almost every aspect of scientific collaboration can be reliably tracked by analyzing coauthorship networks. Nowadays, many co-authorship Serenko and Bontis [2] have developed a global ranking of 20 knowledge management and intellectual capital academic journals through a survey of 233 active researchers in this field. Their goal was to select only academic KM/IC journals focusing on managerial issues, therefore pure technology-centered journals or trade journals were excluded [2]. Since no discrepancies of rank-order were found in the list of top 6 journals while these two scholars applied different rank-order and factor scoring methods to produce the results, we selected the Top 6 journals ranked by Serenko & Bontis as data source of this paper. We collected all the available academic papers published in these top 6 journals from 2000 to 2010 in Proquest Database, as listed in Table 1. Editorials, book reviews, and interviews were excluded from the analysis. In total, 1831 papers with 2471 authors were gathered. The co-authorship network documents scientific collaboration through published articles, where nodes are researchers and a link represents the fact that two researchers have written at least one paper together. In this article, we build a traditional undirected, binary coauthorship network without considering the weight of edges, mainly to simplify the analysis procedures. Only in the connected sub-graphs, weight of edges is taken into consideration. Finally, we build a co-authorship network with 2471 nodes and 2665 edges. TABLE I OVERVIEW OF THE DATA SOURCE FROM 2000 TO 2010 IN PROQUEST DATABASE Num. of Papers Journal Available Period Journal of Knowledge Management (JKM) 586 2000-2010 Journal of Intellectual Capital (JIC) 375 2000-2010 240 2003-2010 65 2007-2009 The Learning Organization (TLO) 271 2000-2010 Knowledge (KPM) 294 2000-2010 Knowledge Management Research & Practice (KMRP) International Journal of Knowledge Management (IJKM) and Process Management Total 1831 V. METRICS FOR THE CO-AUTHORSHIP NETWORK A. Degree Distribution Analysis Node degree is the number of ties linked to other nodes in the network. Network density is the proportion of ties in a network relative to the total number possible [10]. Network density calculated by Ucinet, only 0.0006, indicates that co-authorship network is quite a sparse network. The average node degree is 2.52, namely one researcher cooperate with 2.52 researchers on average. Specifically, the best connected researcher, Bontis, N. has a degree of 38, and the least connected researcher’s degree is 0, which suggests that he/she didn’t cooperate with anyone. In total, there are 356 researchers who prefer to work alone, accounting for 14.40% of the 2471 researchers. A quantity that has been much studied lately for various networks is the degree distribution, P(k), giving the probability that a randomly selected node has k links. Networks for which P(k) has a power-law tail, are known as scale-free networks. The degree distributions of coauthorship network indicate that it is scale-free. The power-law tail is evident from the logarithmically binned data, which suggests that a few researchers have quite a large cooperation network, and in contrast a large sum of common researchers just have a few cooperators. The regression equation is log (f) =-2.694log (N+1) +3.629, R Square=0.8712 (f is the number of nodes that has N links. B. APL and Clustering Coefficient Analysis Path length is the distance of shortest path between a pair of nodes in the network, and average path-length (APL) is the average of these path lengths among all pairs of nodes. Clustering coefficient ranges between 0 and 1, and a higher clustering coefficient indicates a greater ‘cliquishness’ [10]. For all the pairs of nodes that can be reached from each other, the APL is only 2.37. Clearly the coauthorship network is not a connected graph, however, there may exist several connected sub-graphs. Clustering coefficient of co-authorship network is 0.882, indicating a highly short range clustering. Specifically, there are 1122 nodes whose clustering coefficient are 1, accounting for 45.40% of the whole network, and all these nodes have a relatively low degree, ranging from 2-5. It means these researchers usually have a small cooperation network, namely 2-5 cooperators. C. Centrality Analysis At the individual level, one dimension of status in the network can be captured through centrality. Central researchers are well connected to other researchers and metrics of centrality will therefore attempt to measure a researcher’s degree (degree centrality), average distance to all other researchers (closeness centrality), or the degree to which geodesic paths between any pair of researchers passes through the researcher (betweenness centrality). So the most connected researchers can be identified by these three kinds of centrality from different aspects. 1) The 20 Most Central Researchers: The three kinds of 20 most central researchers are listed in Table 2. We highlight those authors (i.e. Bontis, N.; O'Donnell, D. and Voelpel, Sven C.) with italic style and grey background ranked in all three kinds of Top 20 most central researchers; obviously, they are the most connected researchers. 2) The Comparison between Productivity Ranking and Centrality Ranking: Serenko and Bontis [5] conduct a meta-review analysis of the knowledge management and intellectual capital literatures by investigating research productivity. In this article, the top 20 productively researchers are put in the right column of Table 2. As shown in Table 2, except for two researchers (i.e. Bontis, N. and Guthrie, J.) highlighted with bold style in Top 20 productivity which are also listed in Top 20 of centrality ranking, there is an obvious difference between productivity ranking and centrality ranking. On one hand, these big changes are caused by the different periods of data source. Serenko and Bontis[5] collected papers in Top 3 journals before 2003, while in this article papers ranging from 2000-2010 in Top 6 journals are included. Some of these researchers have published a large sum of papers before 2000 since they began their research in KM area as early as 1980s. However, these works are not contained in our article. On the other hand, these highly productive researchers maybe prefer to work alone. Like Rowley, J., ranked 6 in Top 20 productivity, she has only one cooperator in the network, namely her degree is only 1. Hence, she is not in Top 20 of centrality ranking. TABLE II THREE KINDS OF THE 20 MOST CENTRAL RESEARCHERS AND THE 20 MOST PRODUCTIVITY RESEARCHERS No Degree Centrality Closeness Centrality Betweenness Centrality Productivity[4] 1 Bontis, N. Bontis, N. Bontis, N. Bhatt, D. 2 Bukh, N. Andreou, Andreas N. Bontis, N. 3 Guthrie, J. O'Donnell, D. Henriksen, Lars B. Andreou, Andreas N. Kennedy, T. Voelpel, Sven C. Shariq, Syed Z. Stankosky, M. Joia, Luiz A. Guthrie, J. Pablos, Patricia O. O'Regan, P. O'Donnell, D. Rowley, J. Serenko, A. Cleary, P. Bukh, Per N. Wiig, Karl M. O'Donnell, D. Lytras, Miltiadis D. Hannigan, A. 10 Mentzas, G. Seleim, A. Henriksen, Lars B. Davenport, Thomas H. Ribiere, Vincent M. 11 Johansen, M. R. Serenko, A. Johanson, U. KoracKakabadse, A. 12 Psarras, J. Bart, Christopher K. Edwards, John S. KoracKakabadse, N. Hardie, T. Kianto, A. Newman, V. Lönnqvist, A. Skok, W. Carlucci, D. Caddy. I. Sánchez, M. P. Carrillo, J. Seleim, A. Guthrie, J. Prusak, L. Allee, A. Schiuma, G. Beijerse, R. Bart, Christopher K. Beveren, V. Per Metaxiotis, K. Mouritsen, J. Ergazakis, K. 4 5 6 7 8 9 13 14 15 16 17 Stankosky, M. Klein, Jonathan H. Johanson, U. Voelpel, Sven C. Choi, Chong J. 18 Farrell, J. 19 Gwyn, B. 20 McDonald, J. R. Tracey, M. Booker, Lorne D. Sadeddin, K. Voelpel, Sven C. Grant, J. Keow, William Chua C. Richardson, S. Stovel, M. Ganesh A. Analysis of Collaboration Characteristics McAdam, R. Liebowitz, J. Blosch, M. John From these researchers we learn that some highly productive researchers prefer to work alone rather than collaboration, which helps to maintain their original. This is also because they are able to publish papers on these top journals independently. In comparison, team work or collaboration is a better choice for some common researchers if their goal is the top journals. VI. CONNECTED SUB-GRAPHS The co-authorship network is not all connected, which includes many small connected sub-graphs. These subgraphs, to some degree, are on behalf of different kinds of academic groups. We use Pajek to divide the network into connected sub-graphs. It is not hard to find that, most subgraphs have 2-5 actors, there are 6 sub-graphs have more than 15 actors and only 2 sub-graphs whose numbers of actors are more than 30 (with 39 and 57 actors respectively). Next, we only analyze the biggest connected sub-graph for the limitation of pages. The biggest connected sub-graph that extracted by Ucinet have 57 actors, including 47 articles, so we name it 57-Actors Graph. 1) Collaboration Frequency: The collaboration times between the two co-authors include 1, 2, 4, 9 in 57-Actors Graph. Most cooperation in this academic group is in a low frequency (1-2 times), but meanwhile there is a little of high frequent cooperation, for instance, there are 4 times cooperation between Bontis, N. and Bart, Christopher K.; between Stankosky, M. and Mohamed, Mirghani S.; and 9 times between Bontis, N. and Serenko, A.. 2) Geographic Distribution: The paper collects the nationality/district information of all the researchers in the 57-Actors Graph. Most of them are American (16) researchers, next are Canada (12), then Ireland (7), Germany (4), Egypt (2), Malaysia (2), Netherlands (2), Switzerland (2), and other countries/districts, each of which has one researcher. To understand the relationship between researchers` cooperation and nationality distribution, we build a nationality distribution network which is similar to the coauthorship network. The only difference between the two networks is that the edge in the nationality distribution network represents that the two actors are in the same nationality, while the edge in the co-authorship network indicates that they have a collaboration relationship. Then we use QAP (quadratic assignment procedures) method to exam the correlations between the two networks. QAP method is mainly used to analyze the correlations between different networks. The result shows that the correlation coefficient calculated by simple matching algorithm is 0.871, which means that if the value of a unit in the nationality network is 1, then there will be a possibility of 87.1% that the value of the corresponding unit in the co-authorship network is 1. In other words, there is a possibility of 87.1% that-two co-authors in the certain co-authorship network are from the same country. As the significance value is 0.000 (<0.001), which indicates that the significance level is very high [10], the hypothesis that the two networks are highly correlated is established. This shows that researchers who co-publish an article are more likely to come from the same country. 3) Journals for Publication: These 47 articles are not distributed evenly in the 6 journals. About 74.5% of the articles are published on the JKM and JIC. The paper identifies 13 core researchers from the 57-Actors Graph whose published papers cover the whole 47 articles, including 6 core super researchers whose degree numbers are more than 5, they are by descending order: Bontis, N. (38), O'Donnell, D. (15), Stankosky, M. (12), Voelpel, Sven C. (11), and Ribière, Vincent M. (9), Davenport, Thomas H. (6). below, analysis of Bontis’s ego network in next Part), similarly, the group also has a wide range of cooperation, including Europe, America, Asia and Oceania, etc. However, there is no intensive trend about researchers` nationality of Group 3, where 1/3 are from Germany, 1/4 are from USA, and others are from some countries in Europe and Africa. C. Analysis of Bontis`s Ego Network B. Analysis of Divided Groups To better analyze the research difference of different academic groups in the 57-Actors Graph, we divide the graph into 3 small groups according to the tightness of cooperation. According to the division result, we call them Group 1, Group 2 and Group 3 successively. The features of edges (articles) and actors (researchers) in the groups are shown in Table 3. TABLE III THREE KINDS OF THE 20 MOST CENTRAL RESEARCHERS AND THE 20 MOST PRODUCTIVITY RESEARCHERS Articles Group Group 1 (14 articles; 17 authors) Group 2 (28 articles; 28 authors) Group 3 (11 articles; 12 authors) Authors Journal Num Proport ion JKM JIC IJKM KMRP 8 3 2 1 57.1% 21.4% 14.3% 7.1% JIC KPM JKM IJKM TLO 18 4 4 1 1 64.3% 14.3% 14.3% 3.6% 3.6% JIC JKM KPM KMRP 6 3 1 1 54.5% 27.2% 9.1% 9.1% Country/Distr ict USA Norway UK Republic of Korea Slovenia Thailand Canada Ireland Egypt Malaysia Others Germany USA Netherlands Switzerland South Africa 12 1 1 1 Propor tion 70.6% 5.9% 5.9% 5.9% 1 1 12 7 2 2 5 4 3 2 2 1 5.9% 5.9% 42.9% 25.0% 7.1% 7.1% 17.9% 33.3% 25.0% 16.7% 16.7% 8.3% Num From Table 3 we can find that the three groups have their own preferences, there is much difference among their journals where their articles are published and among their nationalities. The co-work articles of Group 1 are mostly published on the JKM, while those from Group 2 focus on the JIC, and Group 3 are also tend to publish their articles on the JIC with a 50% of the total respectively. It is also easy to discover that, most researchers in Group 1 come from USA, and there is respectively one researcher from the country of Norway, UK, Republic of Korea, Slovenia and Thailand, which, indicates that this group has a wide range of cooperation, distributed in America, Europe and Asia. A majority of researchers in Group 2 are with the nationalities of Canada and Ireland (As there is not too much difference between Group 2 and Bontis’s Ego Network, detail analysis may be referred to Nick Bontis is a Canadian academic and the actor with the maximum degrees in the co-authorship network, which means if we only select one scholar as the best connected researcher, he is the only one in KM field undoubtly. He is a Canadian academic at the DeGroote School of Business, McMaster University in Hamilton. He specializes in intellectual capital, knowledge management and organizational learning. We extract Bontis`s ego network to analyze its research focus and evolving trends of the academic group which centers on him. 1) Members of the Bontis-centered Academic Group: Bontis has 23 co-authorships in total, among which 9 are from the same country as him, or even the same university, McMaster University. Except from those Canadian co-authors, Bontis also has much international collaboration. What is worth noticing is that he has as many as 6 co-researchers from Ireland. Besides, he has built some cooperation relationships with researchers that come from USA, Denmark, India, Malaysia and Taiwan. Among those relationships, Serenko, A. has the most frequent collaboration with Bontis; they together have published 9 papers. Next is Bart, Christopher K., who has a number of 5 co-written articles with Bontis. Obviously, their cooperation relations are very stable. 2) Research Focus of the Bontis-centered Academic Group: The Bontis-centered academic group consists of Bontis and his co-authors. He cooperated with two Malaysian researchers in 2000 and one Canadian in 2001. As the best connected researcher in the field of KM, Bontis has gained much multinational cooperation experience. Before 2006, his co-authors each year were mostly from the same country. However, he began to cooperate with multiple nations after 2006, with coauthors in many different countries and districts, including Ireland, Taiwan, USA, Austria, Egypt, etc. In general, many of his cooperation types are transnational ones, and as the time passed by, evolve from one country to multiple countries. According to the introduction on Bontis`s homepage (www.bontis.com), he always uses method of case study. He is good at studying Knowledge Assets such as management strategy of human capital, and concentrates on knowledge assets` influence on organizations. By counting the time and keywords of their co-published papers, analyzing the content change during the session, it is not difficult to find that the group`s research topic always focuses on knowledge management and intellectual capital. From studying some core problems like customer capital and human capital in the enterprises in the early time (2000), step by step, he transferred his research direction to informationization in 2001; Moreover, he turned to study some extension problems such as the relation between employment relationship, leader capability, corporate governance, knowledge management and intellectual capital in the organizations in 2002-2007; Recently in 2009, he began to pay attention to some expanding problems, for example, the application of knowledge management in teaching; Besides, according to Bontis`s articles that published in 2010, we could find that as the technology grows, he put more emphasis on combining knowledge management with digital products, mobile economy, and doing some research on technology acceptance and perceived value. VII. CONCLUSION AND FUTURE WORKS Based on 1831 articles published on the 6 top journals in the knowledge management field from 2000-2010, this study builds a co-authorship network which has 2471 actors and 2665 edges according to the collaboration relationship among the researchers. By analyzing three kinds of network centrality, we respectively identify 20 authors who have the most connections with others according to degree centrality, who are in the core position according to closeness centrality and who play the “bridge” roles in the network according to betweenness centrality, those three authors (i.e. Bontis, N.; O'Donnell, D. and Voelpel, Sven C.) ranked in all three kinds of Top 20 most central researchers obviously are the most connected researchers. From the coauthorship network, it is easy to find Nick Bontis is the actor with the maximum degrees, which means if only one scholar should be selected as the best connected researcher, he is this only one in KM field. Besides, we extract the biggest connected sub-graphs from the network, that is, the 57-Actors Graph, and analyze in detail by studying its cooperation frequency, geographic distribution and journals distribution. Based on the findings, several important implications emerged that warrant discussion. Firstly, there exist a few researchers who have quite a large cooperation network, and in contrast a large sum of common researchers who just have a few cooperators. Then academic researchers usually have a small cooperation network, namely 2-5 cooperators, and there are few connections with researchers outside their cooperation network. However, data shows that some highly productive researchers prefer to work alone rather than collaboration, which helps to maintain their original. This is also because they are able to publish papers on these top journals independently. In comparison, team work or collaboration is a better choice for some common researchers if their goal is the top journals. Still, there are some limitations in our study. Though we have calculated the frequency of the collaboration in the biggest connected sub-graph, we should consider the co-authors sort in the same paper. Moreover, only Top 6 journals are selected in this article at present, which made the sample of papers still small for social network analysis. In the future, more high quality journals should be selected into the data source and made some analysis after considering the co-authors sort and adding the weight information of the whole network. ACKNOWLEDGMENT This work was supported in part by National Natural Science Foundation of China under Grant 70971130, part by Beijing Natural Science Foundation under Grant 9112009, part by Program for New Century Excellent Talents in University, and part by the Fundamental Research Funds for the Central Universities, and the Research Funds of Renmin University of China. REFERENCES [1] Ma, Z., and Yu, K. H. 2010. “Research Paradigms of Contemporary Knowledge Management Studies: 1998– 2007,” Journal of Knowledge Management (14:2), pp. 175189. [2] Serenko, A., and Bontis, N. 2009. “Global Ranking of Knowledge Management and Intellectual Capital Academic Journals,” Journal of Knowledge Management (13:1), pp. 4-15. [3] Serenko, A., Bontis, N., Booker, L., Sadeddin, K., and Hardie, T. 2010. “A Scientometric Analysis of Knowledge Management and Intellectual Capital Academic Literature (1994–2008),” Journal of Knowledge Management (14:1), pp. 3-23. [4] Serenko, A., Cox, R., Bontis, N., and Booker, L. 2011. “The Superstar Phenomenon in the Knowledge Management and Intellectual Capital Academic Discipline,” Journal of Informetrics (5:3), pp.333-345. [5] Serenko, A., and Bontis, N. 2004. “Meta-Review of Knowledge Management and Intellectual Capital Literature: Citation Impact and Research Productivity Rankings,” Knowledge and Process Management (11:3), pp. 185-198. [6] Liu, X., Bollen, J., Nelson, M. L., and Van de Sompel, H. 2005. “Co-authorship Networks in the Digital Library Research Community,” Information Processing & Management (41:6), pp. 1462-1480. [7] Ding, Y. 2011. “Scientific Collaboration and Endorsement: Network Analysis of Co-authorship and Citation Networks,” Journal of Informetrics (5:1), pp. 187-203. [8] Kai, F., Johannes, P., and Detlef, S. 2011. “Co-authorship networks in electronic markets research,” Electronic Markets (21:1), pp. 19-40. [9] Morel, C.M., Serruya, S.J., Penna, G.O. and Guimarães, R. 2009. “Co-authorship Network Analysis: A Powerful Tool for Strategic Planning of Research, Development and Capacity Building Programs on Neglected Diseases,” PLoS Neglected Tropical Diseases (3:8). pp. 1-7. [10] Scott, J. 2005. Social network analysis: a handbook, SAGE Publications Inc.