Healthcare and Science thomsonreuters.com

I started to study bibliometrics when I was a Ph.D. student. In those days, my motivation was to use the citation network approach for my own research.  I was studying material sciences and I noticed that some models are proposed by some research groups but others proposed different models, and that they were not connected in the literature. After I got my Ph.D., I started to think this is definitely common in research and information science.

So, we developed a methodology to analyze the citation networks. And simultaneously, we implemented this technique in web-based software for students and also for policymakers. Here at the University of Tokyo, I teach a class in the Technology Innovation for Management Department. I lecture the students about the citation network approach and introduce our tools and maps, and I show them how to use these for their own research.  And now we have started to offer our tools and maps to policymakers. We have brought together policymakers and researchers to study innovation by exploring science and technology roadmaps. The visualization of the citation network is a very powerful tool.

NEDO (New Energy and Industrial Technology Development Organization) is one funding agency of the Japanese government that has started employing our roadmaps. When a researcher applies for a grant to NEDO, each is encouraged to describe where their research would fit into its technology roadmap. Currently, this is not required, but NEDO strongly recommends it.

The utility of a science map for policymaking has both strong points and weak points. The strong point is it is easy to create a consensus, because it seems to be objective. But I am a bibliometrician, so I know it is not purely objective. The results vary from one methodology to another. That is the weak point. It is a great challenge to confirm the robustness of an analysis. Without confirmation, the reliability of the science map is questionable. Some bibliometricians are now trying to define a standard approach for citation network analysis as a benchmarking tool. For that purpose, we need a basic science of bibliometrics.

The method that has been used in making science maps was based on co-citation clustering.  But our recent study revealed that direct citation is more reliable than traditional approaches such as co-citation and bibliographic coupling. In visualization of science maps, some potential is assumed between the papers. We can assume a number of potentials, but currently we use a locally optimized spring algorithm to ensure the scalability of it. Before the visualization, we cluster the nodes using the Newman-Girvan method. We use this method because we do not need to decide on any threshold in advance, and, second, because it can deal with a large number of entities or nodes. In many cases, we deal with more than a thousand papers where citation analysis can work well. The exploration of better bibliometric approach needs basic understanding of bibliometric data.

Our methods depend somewhat on the purpose of analysis and the granularity needed. To understand the overall structure, we do not need to understand the individual citations. But to understand the individual citation connections the maximum number of nodes represented in the map is around 500 or so, in my experience. The human brain is limited. It cannot comprehend everything at the same time. So, according to the purpose or the granularity of analysis needed, we have a number of possible approaches. One is to bridge the nodes and links and another is to sum up the information by using clustering. And then there is the possibility of timeline visualizations - dynamic depictions over time.

The data on funding acknowledgements recently added to the Web of Science is a great contribution of Thomson Reuters. These data will help the bibliometrician provide useful evidence to policymakers about their funding decisions. We have not yet used these data but I think it is a future direction of our research community.