Abstract - Analyses by countries are in high demand, because many policy-makers and business-planners think by country. We are providing the system of citation network analysis, and we strengthen analyses by countries in this work. So, user will be able to easily extract country’s keywords and compare text-based similarity and number of international co-authors between countries. These new functions would unveil relations between countries. The purpose of this work is to establish a methodology to detect potential collaborative and competitive countries in science and technology, comparing text-based similarity and number of international co-authors. In this work, we focused on analysis about Japan as an example, and we could pick up several implicit relations which have potential to grow collaborative or competitive relations. Keywords - bibliometrics, citation network, text-based similarity, international co-author I. INTRODUCTION To survive worldwide competitions of research and development in the current rapid increase of information, decision-makers and researchers need to be supported to find promising research fields and papers. It becomes difficult to find a bit of beneficial data in too much heavy flood of information. We have developed “Academic Landscape System” [1] to analyze citation network of papers or patents, and it provides citation network and cluster analyses. Based on our experience in providing the system, some users of the system demand to analyze with countries' information, though such analyses do not be provided yet. Analyses by countries are in high demand, because many policy-makers and business-planner think by country from our several meetings with them. The purpose of this work is to establish a methodology to detect potential collaborative and competitive countries in science and technology by country-level analysis. We suggest gaps of text-based similarities and numbers of co-authors between countries. Text-based similarities mean both of explicit and implicit connections, and numbers of co-authors mean only explicit connections. Thus, the trend gaps between text- based similarities and numbers of co-authors will show implicit relations. II. METHODOLOGY After clusters of papers are identified and selected on the basis of Japan’s share of papers in each clusters, text- based similarities and numbers of co-authors between countries are calculated in the selected clusters. A. Citation Network Analysis and Identifying Clusters Garfield [2] began the citation network analysis, and it became an efficient tool to extract popular topics and important papers [3]. Topological measures in citation networks of scientific publications [4] proposed a methodology for detecting emerging researches using temporal change and relations between papers. Several papers present perspective and geometric science maps with citation networks [5][6]. In the citation networks, a node is defined as a paper, and an edge is defined as a citation relation between papers. The citation networks reflect authors' thought about the contents of other papers related to the authors' paper. In this paper, we perform the network analysis with the following five steps. In the step (1), the bibliographic information of papers is collected, and in the step (2), citation networks of direct citations are constructed. The using relationship is the direct citation between the citing paper and the cited paper. The direct citations has less amount of calculation and describes relationships more clearly than co-citation and bibliographic coupling, though the direct citation has a flaw that the relationships published simultaneously are not available. In the step (3), only the largest graph component of the citation networks is used, because this paper focuses on the relationship among the papers, and we should therefore eliminate the papers that have no citation from or to any others. After extracting the largest connected component, in step (4), the network is divided into clusters where papers are densely connected by citations, with a topological clustering method. A fast clustering algorithm developed by Newman [7] was used for clustering. In the step (5) after clustering, topics of each cluster are identified from keywords and abstract of papers. A sequence of the procedure between step (2) and (4) is run at the Academic Landscape System [1] after the bibliographic information of papers is inputted into it. On the step (1), papers of the top 0.1% number of citations are derived from all the papers published between 2004 and 2014 using Science Citation Index Expanded (SCI-EXPANDED), Social Sciences Citation Index (SSCI) and Arts & Humanities Citation Index (A&HCI) updated 2014-03-07, which are provided as a service of academic papers' database by Thomson Reuters. 16,626 papers are retrieved, and their bibliographic records are used for cluster analysis. For the largest graph component, 14,704 papers are divided into 79 clusters. Bibliometric Methodology to Detect Collaborative and Competitive Countries S. Iwami 1 , F. Tacoa 1 , J. Mori 1,2 , Y. Kajikawa 1,3 , I. Sakata 1,2 1 Innovation Policy Research Center, the University of Tokyo, Tokyo, Japan 2 Department of Technology Management for Innovation, the University of Tokyo, Tokyo, Japan 3 Graduate School of Innovation Management, Tokyo Institute of Technology, Tokyo, Japan ({iwami, ftacoa, jmori, isakata}@ipr-ctr.t.u-tokyo.ac.jp, kajikawa@mot.titech.ac.jp)