Knowledge Map of Publications in Research Policy 1

Hsin-Ning Su1, Pei-Chun Lee1,2

Science and Technology Policy Research and Information Center, National Applied Research laboratories, Taipei, Taiwan 2 Graduate Institute of Technology and Innovation Management, National Cheng Chi University, Taipei, Taiwan Abstract--Research Policy as a leading journal among others in the fields of social science, plays as an important platform for policy researchers to have their research results published, documented and shared internationally. This study uses Research Policy as a window or an indicator to understand overview of global Sci-Tech and Innovation Policy/Management researches, as well as unveil how papers in Research Policy are correlated to each other and how quantitative technology management can be possibly obtained. This study positions research foci in Research Policy by keyword-based network analysis. A keyword-based network which is also named as “research focus parallelship network” can be visually obtained and network properties, e.g. degree centrality, betweenness centrality, closeness centrality, can be calculated. The keyword-based network can be depicted differently to reflect its research focus parallelship as well as knowledge linkage implication by choosing different information as network actors such as keyword, first author, institute, or country. Totally 2014 keywords contained in 934 papers published in Research Policy in the period of 1998-2008.

I. INTRODUCTION Kuhn published “The structure of scientific revolution” in 1970 and popularized the terms “paradigm” and “paradigm shift” [1]. Dosi investigated technology trajectory on the basis of paradigm shift and found continuous innovation can be regarded as proceeding of technology paradigm, while discontinuous innovation might be the initiation of a new paradigm [2]. The differentiation between continuous innovation and discontinuous innovation may be positive for understanding initiation of a new paradigm as well as position and diffusion of a specific technology or knowledge. A lot of methodologies have been proposed and applied into various knowledge fields for understanding or mapping their paradigms. However, what usually used for this purpose is bibliometric analysis on patents or scientific papers. Co-citation based method is widely used as a way in bibliometric analysis, e.g. author co-citation and journal cocitation [3], to uncover the tracks of knowledge evolution in the society. But this citation-based method only presents information about relations among literatures without providing any clue of what the core concept of each individual actor, i.e. paper or patent, etc., in knowledge evolution is about [4]. Therefore, the co-word analysis is developed to calculate frequency of co-keywords among literatures, which provides another way of mapping knowledge evolution with disclosure of condensed core concept by meaningful each keywords [5].

A lot of attempts have been made to explore way of mapping knowledge evolution. Keyword based analysis as a type of co-word analysis [6][7] started to play an important role in understanding the dynamics of knowledge development [8]. Ding et al. mapped information retrieval research by using co-word analysis on papers collected from Science Citation Index (SCI) and Social Science Citation Index (SSCI) for the period of 1987–1997 [4]. Baldwin et al. mapped ethics and dementia research by using keywords [9]. Tian et al. [10] used Institute for Science Information (ISI) database to measure scientific output of the field of (Geographic Information System) GIS by using keywords [10]. Similar approaches have been made to map knowledge evolution in other fields, such as software engineering [11], chemistry [5], scientometrics [12], neural network research [13][14], biological safety [15], optomechatronics [13], bioeletronics [16], adverse drug reactions [17][18], biotechnology [19][20], environmental science [21], Condensed Matter Physics [22],etc., and even this keywordbased analysis has been applied to investigation of a phenomenon or a more specific topic such as severe acute respiratory syndrome (SARS) [23]and tsunami [24], Parkinson’s disease [25]. Some analysis methods are enhanced by combing keyword analysis. For example, morphology analysis is a conventional method of forecasting future technology and identifying technology opportunities. Yoon and Park (2005) argued that morphology analysis is subject to limitations because there is no scientific or systematic way in establishing the morphology of technology [26]. Therefore, keyword-based morphology analysis that is supported by a systematic procedure and quantitative data for concluding the morphology of technology is proposed. An example of thin film transistor–liquid crystal display (TFT-LCD) is studied to illustrate the detailed procedure of this keyword-based morphology analysis. II. MAPPING KNOWLEDGE OF TECHNOLOGY FORESIGHT BY KEYWORD NETOWORK A. Combination of Keywords and network theory The basic components of a social network can be different forms of social actors, for example, humans, organizations, countries. A social network formed on the basis of social exchange can be used for understanding how resources are exchanged in this network, how social actors are positioned to influence resource exchange, and which resource exchange

is important [27][28][29]. Each of resource exchange is a social network relation or a “tie” maintained by social actors at both end of the “tie”, the strength of a tie is a function of the number of resource exchange, the type of exchange, the frequency of resource exchange, or even how close the two connected actors are [30]. Social network analysis is an interdisciplinary research field, Granovertter [31][32] proposed the theory of weak tie after his social network research. Granovetter surveyed 282 workers in total, in regards to the type of ties between the job changer and the contact person who provided the necessary information. Of those who found jobs through personal contacts, only 16.7% reported seeing their contact often. This illustrates social network analysis is a proxy which provides interconnection between microscopic analysis and macroscopic analysis. In the late 1990s, collaboration between researchers from different fields by the use of social network analysis had been initiated so social network analysis become more interdisciplinary. At the time, Barabasi and Albert [33] demonstrate that the algebraic distribution in the connectivity of scale-free network is caused by two basic factors in the temporal evolution of the network: growth and preferential attachment [33]. Watts and Strogatz [34] published a breakthrough paper in the journal- Nature [34], and a book entitled ”Six Degrees: The Science of A Connected Age” [35], together with other interdisciplinary works contribute to expansion of small world concept from conventional neuro-science and bio-information system to any natural or human system that can be modeled by network. Social network analysis based on keyword has also been explored, Motter et al. [36] constructed a conceptual network from the entries in a thesaurus dictionary and consider two words connected if they express similar concepts. He argued that language network exhibits the small-world property as a result of natural optimization and these findings are important not only for linguistics, but also for cognitive science. Marshakova-Shaikevich [37] attempted to build a semantic map of a field of women’s studies by document clustering on the basis of lexical similarity of titles and word clustering on the basis of co-occurrence of words in the same documents. Accordingly, the purpose of this research is to shed light on combination of social network analysis and bibliometric analysis on publications in Research Policy by using different publication information, e.g. keyword, research institute, origin of country, as actors in network. The network actors and linkages corresponded to publication information and keyword occurrence, respectively, can be visualized and thus dynamic knowledge evolution can be mapped out. Furthermore, network properties for keyword-based network development in this study can be calculated to obtain qualitative analysis on knowledge evolution [32]. B. Construction knowledge map of Research Policy Research and development, technology, and innovation managements are increasingly important for the industry,

governments, academies, research institutes. Optimal policies are always required for sustaining stable development-based R&D, technology and innovation managemet. Research Policy journal plays as an extremely platform for policy researchers to have their research results published, documented and shared internationally. Research Policy as a leading journal among others in the fields of economics, law, political science, and other social science journals, is evaluated as the top journal with high impact factors and ranked the 8th among the world's top journals in the Management and the 2nd in the Planning & Development category [38]. As stated in Research Policy [38] - Research and development (R&D) and innovation today absorb very considerable resources. These activities have great influence on the policies of industrial firms, government departments, universities and nations. Research Policy is a multidisciplinary journal devoted to the policy and management problems posed by innovation, R&D, technology and science, and related activities concerned with the acquisition of knowledge (learning) and its exploitation. Its papers examine the interaction between these activities and economic, social, political and institutional processes. Many of the papers are empirically focused but others are more theoretical. They are written by both academic analysts and practitioners of R&D and innovation processes. The journal is international in scope and reaches an audience of academics, industrialists, government officials and others interested in these issues. Its leading academic status and influence are reflected in a high 'impact factor' for a social science journal” (Elsevier, 2008). This study originally aims to understand policy and management studies of science, technology and innovation, which directly fit the scope of Research Policy. But due to fact that limited number of papers of one single journal is hard to obtain objective view of global development, this study therefore defines the research scope as “Research Policy Knowledge Map” in stead of “Sci-Tech and Innovation Policy Map”. In other words, it is not attempted in this study to draw a large and detail map for global policy research, but trying to use Research Policy as a window or an indicator to approach global Sci-Tech and Innovation Policy/Management and also answer questions such as which country or which research institute contribute the most to this field, which country located at the core of this field, and how to draw relations of policy researches conducted by different research institutes or countries. III. RESEARCH METHOD This research integrates social network analysis and bibliometric keyword analysis to draw a picture for the development of knowledge of Research Policy journal, or can be called “Research Policy knowledge map” where each country, research institute, or researcher that contributed to Research Policy Journal can be positioned.

A. Preparation of dataset This research selects “Research Policy” journal as research target, trying to map knowledge or technology trajectories of this journal on the basis of author keyword occurrence. However papers published in Research Policy do not necessarily contain author keywords before 1998. Therefore, this study selects all papers published in Research Policy from 1998-2008 and prepares structured datasets of basic paper information. e.g. paper title, abstract, first author, research institute, country, publication time, author keywords in spreadsheet for following analyses. In summary, totally 934 papers with 3923 keywords (averaged 4.2 keywords per paper) are published by 427 institutes from 42 countries from 1998-2008.

B. Keyword revision and basic statistical analysis Due to the fact that different words can be used for describing the same meaning, it is necessary to standardize keywords that used to express the same or similar meaning. For example, 1) bibliometric analysis, bibliometric study are revised as bibliometrics form; 2) technique, technologies, technology are revised as technology; 3) university-industry linkage, university-industry linkages, university-industry R&D links, university-industry R&D relationships are standaridized as university-industry links. After removal of duplication and revision, a total of 2014 author keywords are obtained. Table 1 shows keyword occurrences.

TABLE I. TOP 20 HIGH OCCURRENCE KEYWORDS OF PAPERS IN RESEARCH POLICY BETWEEN 1998-2008 Keyword (Total of 2014) Occurrence Keyword (Total of 2014) Occurrence innovation 148 technological policy 21 R&D 81 collaboration 20 patent 65 innovation policy 19 technological transfer 38 R&D collaboration 19 innovation system 29 absorptive capacity 17 national innovation system 26 multinational firms 16 technology 26 productivity 16 biotechnology 25 university 16 knowledge 23 network 15 entrepreneurship 22 institution 14

C. Keyword network visualization Networking of keyword is based on sufficient relations among keywords. The relation is presented as a “network tie”. This study provides two methods of network tie generation. 1) Research focus parallelship network: a relation between two different papers occurred because these two papers share at least one same keyword. Network generated by this method is defined as “paper network”, If different types of actors are selected, such as selecting research institute by which keywords are shared, “institute network” or “country network” can be generated, respectively. 2) Keyword co-occurrence network: relations among plural keywords occurred because these keywords are listed in the same literature paper. Network generated by this method is defined as “keyword network.” These two methods are further explained as follows: 1) Research focus parallelship network: Relation between two different papers occurred because these two papers share at least one same keyword. For example, paper is used as a network actor (network node) and any of two actors sharing one same keyword will be linked. This is based on an assumption made in this study that keyword represents the core of research of a paper and any two papers sharing same keyword implies these two researches are partially overlapped in an area that can be represented by that keyword.

The two papers are thus defined as a pair of parallel papers and the constructed network is defined as “research focus parallelship network”. However, network node is not necessarily paper, it can also be selected from different actors, e.g. first author, research institute, country, by which papers are published. In this study, first author, research institute, country are selected as network actor to construct research focus parallelship network in order to understand knowledge evolution of Research Policy at micro-, meso-, and macro-levels, respectively. 2) Keyword co-occurrence network: Relations of keywords are formed because keywords are listed in the same paper. Author eeywords listed in the same papers are linked together because they are all terms that can be used to represent the core of a research paper and stronger relations to each other can be expected. Keywords in the same paper share equal importance for the paper. The shortest distance between any two keywords that are not directly linked can be regarded as how close the two keywords are. Keywords with higher network centrality are supposed to be closer to core concept of Research Policy. D. Network properties calculation Computer software is used to visualize research focus parallelship network and keyword co-occurrence network and then network properties are subsequently calculated. In

social network theory, centrality is used to estimate influence of actors. Centrality as an indicator can be used to understand in what degree an actor is able to obtain or control resources. Brass and Burkhardt [39] indicated network centrality is one source of influence from the viewpoint of organizational behavior, a person with higher centrality in an organization is always the one with higher influence. Freeman [40] suggested three methods of centrality measurement for a network: 1) Degree Centrality, 2) Between Centrality, and 3) Closeness Centrality. Network properties are calculated by the above three methods in this study in order to understand the power of influence of first author, research institute, country in the field of Technology Foresight. A social network can be either a directed network or an undirected network, networks constructed in this research are undirected networks because no in-and-out concept, e.g. causal relation, position difference, flow, or diffusion, existed behind any linked keywords. 1) Degree Centrality Network nodes (actor) which directly linked to a specific node are neighborhood of that specific node. The number of neighbors is defined as nodal degree, or degree of connection. Granovetter (1973) suggested nodal degree are proportional to probability of obtaining resource [32]. Nodal degree represents to what degree a node (actor) participates the network, this is a basic concept for measuring centrality. Degree Centrality is defined as the number of direct linkage between actor i and other actor.

d (i ) = ∑ mji j

mij=1 if actor i and actor j are linked 2) Betweenness Centrality The concept of betweenness is a measure of how often an actor is located on the shortest path (geodesic) between other actors in the network. Those actors located on the shortest path between other actors are playing roles of intermediary that help any two actors without direct contact reach each other indirectly. Actors with higher betweenness centrality are those located at the core of the network.

gjik b(i ) = ∑ j , k ≠1 gjk gjk：shortest path between actor j and actor k gjik：the shortest path between actor j and actor k that contains actor i 3) Closeness Centrality

The closeness centrality of an actor is defined by the inverse of the average length of the shortest paths to/from all the other actors in the network. Higher closeness centrality indicates higher influence on other actors. N

1 j =1 dji

c(i) = ∑

dji ：shortest path between actor j and actor i IV. RESULT AND DISCUSSION A. Paper sample analysis Among all the retrieved 934 papers, US is the country with the most papers (184 papers), then UK (161), Netherlands (89), Germany (72), France (68), Italy (58), Spain (44), etc., as shown in Table 2. Totally 42 countries have publication in Research Policy between 1998-2008. TABLE II. COUNTRIES WITH HIGH NUMBER OF PAPERS IN RESEARCH POLICY BETWEEN 1998-2008 Country No. of Paper % (Total of 42) US 184 19.70 UK 161 17.24 Netherlands 89 9.53 Germany 72 7.71 France 68 7.28 Italy 58 6.21 Spain 44 4.71 Canada 32 3.43 Japan 32 3.43 Sweden 19 2.03 Australia 15 1.61 Switzerland 14 1.50 Denmark 13 1.39 Finland 13 1.39 Belgium 12 1.28 Austria 10 1.07 Norway 10 1.07 India 8 0.86 Israel 8 0.86 Singapore 8 0.86

Research institutes by which papers are published are also calculated, the research institutes that publish the most paper in this field is University of Sussex (34 papers), University of Manchester (20), Eindhoven University of Technology (17), Harvard University (16), Georgia Institute of Technology (14), University of California Berkeley (14). A total of 427 research institutes have publication in Research Policy between 1998-2008 and Table 3 lists research institutes that publish equal or more than seven papers.

TABLE III. RESEARCH INSTITUTES WITH HIGH NUMBER OF PAPERS IN RESEARCH POLICY BETWEEN 1998-2008 Research institute (Total of 427) No. of Paper % University of Sussex 34 3.64 University of Manchester 20 2.14 Eindhoven University of Technology 17 1.82 Harvard University 16 1.71 Georgia Institute of Technology 14 1.50 University of California Berkeley 14 1.50 University of Louis Pasteur 12 1.28 Bocconi University 12 1.28 University of Maastricht 12 1.28 University of London 12 1.28 Fraunhofer Institute for Systems and Innovation Research 9 0.96 University of Leiden 8 0.86 University of Warwick 8 0.86 Catholic University of Leuven 7 0.75 Max Planck Institute of Economics 7 0.75 Sant Anna School of Advanced Studies 7 0.75 University of Pavia 7 0.75 University of Tokyo 7 0.75 University of Amsterdam 7 0.75

Among a total of 761 first authors, those who publish equal to or more than five papers are David B. Audretsch (6), Henry Etzkowitz (6), Robert J. W. Tijssen (5), David C. Mowery (5). B. Keyword network analysis Netowrk Visualization Keywords of the 934 papers are used as basis for network construction to obtain research focus parallelship network by

the use of different network actors, e.g. country, research institute, first author (individual paper). 1) Country as network actor in research focus parallelship network: Papers are classified by country, and any two actors (country) with the same keyword are linked together. Totally there are 42 network actors, and 552 network ties. As shown in Figure 1, European countries are the major countries that contribute to this field.

Fig. 1. Country network of Research Policy between 1998-2008

2) Research institute as network actor in research focus parallelship network: Papers are classified by research institute, and any two actors (research institute) with the same keyword are linked

together. A total of 427 network actors and 12880 network ties are obtained and shown in Figure 2.

Fig. 2. Research institute network of Research Policy between 1998-2008

3) First author (individual paper) as network actor in research focus parallelship network: Any two actors (first author/individual paper) with the same keyword are linked together. A total of 934 network

actors and 23040 network ties are obtained and shown in Figure 3

Fig. 3. First author network of Research Policy between 1998-2008

4) Keyword as network actor in keyword co-occurrence network: Each author keyword is treated as a network actor, keywords within the same papers are linked together. A total

of 2014 network actors and 2864 network ties are obtained are shown in Figure 4.

Fig .4. Keyword network of Research Policy between 1998-2008

Netowrk properties calculation Calculation methods mentioned previously are used to calculate network properties for networks shown in Figure 14, i.e. Between Centrality, Degree Centrality, Closeness Centrality, to understand their centralities. For research focus parallelship network with country as network actor (Figure 1), countries with top ten network properties are listed in Table 4. UK has the highest centrality and then France, Germany, Italy, etc. The number of papers that each country contributes to this field is different, but it is easily anticipated that country with more papers tends to have

more linkages to other countries by their larger number of papers. It is therefore observed that countries with more papers have higher centrality and are thus positioned at the core of the network. Countries with more papers shown in Table 2 are a bit consistent to countries with higher centrality calculated in Table. 4. However, the number of papers for US is ranked No. 1, but its centralities is ranked No. 4-6. European countries seems to perform better than the US in terms of centralities. Japan is the only Asian nation that is positioned in the top 10 countries (No. 9 in terms of number of papers and No. 10 in degree centrality).

TABLE IV. TOP 10 COUNTRIES WITH HIGHEST NETWORK PROPERTIES IN RESEARCH POLICY BETWEEN 1998-2008 Ranking Degree Centrality Betweenness Centrality Closeness Centrality 1 UK Italy UK 2 France Germany Italy 3 Germany China Germany 4 Italy UK US 5 Spain US Spain 6 US Spain France 7 Netherlands France Netherlands 8 Canada Netherlands Canada 9 Sweden Canada Sweden 10 Japan Norway Norway

For research focus parallelship network with research institute as network actor (Figure 2), research institute with top ten network properties are listed in Table 5. Research institute with the highest centralities is University of Sussex. Similar to the previous observation that countries with more papers have higher centralities, research institute with more papers also tends to have higher centralities. Research

institutes with more papers shown in Table 3 are consistent to those with higher centralities calculated in Table 5. Research institutes with higher centralities in Table 5 are all European organization except University of California Berkeley, Harvard University, and George Institute of Technology that are all from the US.

TABLE V. TOP 10 RESEARCH INSTITUTES WITH HIGHEST NETWORK PROPERTIES IN RESEARCH POLICY BETWEEN 1998-2008 Ranking Degree Centrality Betweenness Centrality Closeness Centrality 1 University of Sussex University of Sussex University of Sussex University of California Eindhoven University of Technology University of California Berkeley 2 Berkeley Fraunhofer Institute for Systems and Fraunhofer Institute for Systems Harvard University 3 Innovation Research and Innovation Research 4 University of Maastricht University of Manchester University of Maastricht 5 Bocconi University University of California Berkeley Bocconi University Chalmers University of Fraunhofer Institute for Systems and 6 Chalmers University of Technology Technology Innovation Research 7 University of Manchester Bocconi University University of Oslo 8 University of Oslo University of Leiden University of Manchester 9 Rensselaer Polytechnic Institute University of Maastricht Georgia Institute of Technology 10 Georgia Institute of Technology Georgia Institute of Technology Rensselaer Polytechnic Institute

For keyword co-occurrence network with keyword as network actor (Figure 6), keywords with top 20 centralities are listed in Table 6. Due to the research target set as papers published in Research Policy, these high-centrality author keywords listed in Table 6 are expected to be consistent to the main subjects covered by Research Policy. The main subjects includes competence/capability (e.g. core, dynamic), entrepreneur/entrepreneurship, evolutionary/schumpeterian economics, industrial clusters, innovation management/ policy/strategy, knowledge (creation, transfer, exploitation etc.), system(s) of innovation (national, regional, sectoral

etc.), learning (e.g. organisational) and experimentation, problem-solving, product and process development, R&D management, research and development (R&D), research policy, science policy, technology management/ policy/strategy” [38]. And by combing Table 6 and subjects covered by Research Policy, it is therefore possible to depict a scenario that policy research is strongly related to innovation and is expected to be contributed to innovation system which relies on knowledge, R&D, technology transfer, patents, particularly for the field of biotechnology.

TABLE VI. TOP 10 KEYWORDS WITH HIGHEST NETWORK PROPERTIES IN RESEARCH POLICY BETWEEN 1998-2008 Ranking Degree Centrality Betweenness Centrality Closeness Centrality 1 innovation innovation innovation 2 R&D R&D R&D 3 patent patent patent 4 innovation system biotechnology biotechnology 5 biotechnology innovation system innovation system 6 national innovation system knowledge knowledge 7 knowledge technological transfer Technology 8 technological transfer national innovation system university 9 Technology innovation policy alliance 10 technological policy technology technological change 11 entrepreneurship Intellectual property rights technological policy 12 innovation policy university science 13 university technological policy national innovation system 14 collaboration multinational firms technological transfer 15 R&D collaboration R&D collaboration collaboration 16 venture capital foreign direct investment institution 17 Intellectual property rights institution entrepreneurship

V. CONCLUSION Research Policy as a leading journal in the fields of economics, law, political science, and other social sciences has been providing a way to look into world’s technology and innovation management related researches. Its academic status and influence are reflected in a high impact factor for a social science journal. This study is to understand the overview of researches published in Research Policy in the period of 1998-2008 by social network analysis, a total of 934 papers between 19982008 are retrieved from Research Policy journal. The 934 papers contains 2014 different keywords, and are published by 427 research institutes from 42 countries. The 934 papers with keywords are used for drawing keyword based research parallel network, namely 1) Country as network actor, as shown in Figure 1, 42 network actors, and 552 network ties. 2) Research institute as network actor, as shown in Figure 2, 427 network actors and 12880 network ties, 3) First author (individual paper) as network actor, Figure 3, 934 network actors and 23040 network ties. Also, 4) Keyword as network actor in keyword co-occurrence network, Figure 4, 2014 network actors and 2864 network ties are obtained. Network properties are analyzed on these obtained four networks to obtain Degree Centrality, Betweenness Centrality, Closeness Centrality of network actors. Therefore, a knowledge evolution map, in terms of country, research institute and first author, can be obtained so positioning of each actor in Research Policy can be quantitatively and visually obtained. According to network centrality measurements, Table 4Table 6, countries positioned at the core are European countries. US ranks No. 1 in terms of number of paper but does only rank No. 4-6 in terms of centrality measurement. Japan is the only nation that is in the Top 10 degree centrality country. The sum of papers from US, UK, Netherlands, Germany are more than 50% of total papers, indicating strong policy research capability of these countries. The purpose of this study is to map knowledge for Research Policy by the use of paper retrieved from database. The research results might be a good but not perfect channel for understanding global policy researches, because Research Policy is an English journal without comprising papers in other language, also the total number of paper is only 934 from 1998-2008 and not quite enough for obtaining macroscopic view of global policy research. However, this study provides a channel or an indicator to approach global policy research map, and also demonstrates a quantitative way of mapping knowledge by social network analysis The similarity between linked network actors can be calculated by method such as data-mining, so the obtained similarity can be used as a function of network tie, e.g. thickness of tie, length of tie, and network properties: Degree Centrality, Betweenness Centrality, or Closeness Centrality can be used as network actor properties such as actor's node

size or color, e.g. network actor node size in Figure 1-4 is proportional to its Degree Centrality, to allow more informative visualization. The dynamics of knowledge map obtain in this study can also be investigated by drawing network and calculating network properties for each yeas, so dynamic evolution of actors/knowledge can be quantitatively obtained. REFERENCES [1] [2] [3] [4] [5]

[6] [7] [8] [9] [10] [11] [12] [13]

[14] [15] [16] [17] [18] [19]

T.S. Kuhn, J. Dewey, and O. Neurath, The structure of scientific revolutions, University of Chicago Press Chicago, 1970. G. Dosi, “Technological paradigms and technological trajectories: a suggested interpretation of the determinants and directions of technical change,” Research policy, vol. 11, 1982, pp. 147-162. H. Small, “Co-citation in the scientific literature: A new measure of the relationship between two documents,” Journal of the American Society for information Science, vol. 24, 1973. Y. Ding, G.G. Chowdhury, and S. Foo, “Bibliometric cartography of information retrieval research by using co-word analysis,” Information Processing and Management, vol. 37, 2001, pp. 817-842. M. Callon, J.P. Courtial, and F. Laville, “Co-word analysis as a tool for describing the network of interactions between basic and technological research: The case of polymer chemsitry,” Scientometrics, vol. 22, 1991, pp. 155-205. J. King, “A review of bibliometric and other science indicators and their role in research evaluation,” Journal of Information Science, vol. 13, 1987, p. 261. J. Law and J. Whittaker, “Mapping acidification research: A test of the co-word method,” Scientometrics, vol. 23, 1992, pp. 417-461. B.M. Gupta and S. Bhattacharya, “A bibliometric approach towards mapping the dynamics of science and technology,” DESIDOC Bulletin of Information Technology, vol. 24, 2004, pp. 3-8. C. Baldwin, J. Hughes, T. Hope, R. Jacoby, and S. Ziebland, “Ethics and dementia: mapping the literature by bibliometric analysis,” International Journal of Geriatric Psychiatry, vol. 18, 2003. Y. Tian, C. Wen, and S. Hong, “Global scientific production on GIS research by bibliometric analysis from 1997 to 2006,” Journal of Informetrics, vol. 2, 2008, pp. 65-74. N. Coulter, I. Monarch, and S. Konda, “Software engineering as seen through its research literature: A study in co-word analysis,” Journal of the American Society for Information Science, vol. 49, 1998. J.P. Courtial, “A coword analysis of scientometrics,” Scientometrics, vol. 31, 1994, pp. 251-260. E.C.M. Noyons and A.F.J. Van Raan, “Monitoring scientific developments from a dynamic perspective: Self-organized structuring to map neural network research,” Journal of the American Society for Information Science, vol. 49, 1998. A.F.J. Van Raan and R.J.W. Tijssen, “The neural net of neural network research: an exercise in bibliometric mapping,” Scientometrics(Print), vol. 26, 1993, pp. 169-192. A. Cambrosio, C. Limoges, J.P. Courtial, and F. Laville, “Historical scientometrics? Mapping over 70 years of biological safety research with coword analysis,” Scientometrics, vol. 27, 1993, pp. 119-143. S. Hinze, “Bibliographical cartography of an emerging interdisciplinary discipline: the case of bioelectronics,” Scientometrics, vol. 29, 1994, pp. 353-376. F. Rikken, H.A.L. Kiers, and R. Vos, “Mapping the dynamics of adverse drug reactions in subsequent time periods using INDSCAL,” Scientometrics, vol. 33, 1995, pp. 367-380. A. Clarke, M. Gatineau, M. Thorogood, and N. Wyn-Roberts, “Health promotion research literature in Europe 1995 2005,” The European Journal of Public Health, vol. 17, 2007, p. 24. A. Rip and J.P. Courtial, “Co-word maps of biotechnology: An example of cognitive scientometrics,” Scientometrics, vol. 6, 1984, pp. 381-400.

[20] M.A. De Looze and J. Lemarié, “Corpus relevance through co-word analysis: An application to plant proteints,” Scientometrics, vol. 39, 1997, pp. 267-280. [21] Y.S. Ho, “Bibliometric Analysis of Adsorption Technology in Environmental Science,” Journal OF Environmental Protection Science, vol. 1, 2007, pp. 1-11. [22] S. Bhattacharya and P.K. Basu, “Mapping a research area at the micro level using co-word analysis,” Scientometrics, vol. 43, 1998, pp. 359372. [23] W.T. Chiu, J.S. Huang, and Y.S. Ho, “Bibliometric analysis of Severe Acute Respiratory Syndrome-related research in the beginning stage,” Scientometrics, vol. 61, 2004, pp. 69-77. [24] W.T. Chiu and Y.S. Ho, “Bibliometric analysis of tsunami research,” Scientometrics, vol. 73, 2007, pp. 3-17. [25] T. Li, Y.S. Ho, and C.Y. Li, “Bibliometric analysis on global Parkinson's disease research trends during 1991–2006,” Neuroscience Letters, vol. 441, 2008, pp. 248-252. [26] B. Yoon and Y. Park, “A systematic approach for identifying technology opportunities: Keyword-based morphology analysis,” Technological Forecasting & Social Change, vol. 72, 2005, pp. 145160. [27] N. Nohria, R.G. Eccles, and H.B. School, Networks and organizations: structure, form, and action, Harvard Business School Press Boston, MA, 1992. [28] S. Wasserman and J. Galaskiewicz, Advances in social network analysis: Research in the social and behavioral sciences, Sage, 1994.

[29] B. Wellman and S.D. Berkowitz, “Introduction: Studying social structures,” Social structures: A network approach, 1988, pp. 1–14. [30] P.V. Marsden and K.E. Campbell, “Measuring tie strength,” Social Forces, 1984, pp. 482-501. [31] M.S. Granovetter, “Changing jobs: channels of mobility information in a suburban community,” Unpublished doctoral dissertation, Harvard University, Boston, MA, 1970. [32] M.S. Granovetter, “The strength of weak ties,” American journal of sociology, vol. 78, 1973, p. 1360. [33] A.L. Barabasi and R. Albert, “Emergence of scaling in random networks,” Science, vol. 286, 1999, p. 509. [34] D.J. Watts and S.H. Strogatz, “Collective dynamics ofsmallworld'networks,” Nature, vol. 393, 1998, pp. 440-442. [35] D.J. Watts, Six degrees: The science of a connected age, WW Norton & Company, 2003. [36] A.E. Motter, A.P.S. de Moura, Y.C. Lai, and P. Dasgupta, “Topology of the conceptual network of language,” Science Phys Rev E, vol. 65, 1999, p. 065102. [37] I. Marshakova-Shaikevich, “Bibliometric maps of field of science,” Information Processing and Management, vol. 41, 2005, pp. 15341547. [38] Elsevier, “Elsevier,” Research Policy. [39] D.J. Brass and M.E. Burkhardt, “Centrality and power in organizations,” Networks and organizations: Structure, form and action, vol. 191, 1992, p. 215. [40] L.C. Freeman, “Centrality in social networks: Conceptual clarification,” Social networks, vol. 1, 1979, pp. 215-239.