Decreasing diversity in Japanese science, evidence from in-depth analyses of science maps

Abstract

This study aimed to monitor the status of scientific research in Japan, especially the diversity of scientific research, based on in-depth analyses of science maps. Analyses of six consecutive maps showed decreasing diversity in Japanese science relative to benchmarking countries. We proposed a new concept for a Sci-GEO chart (representing geographical characteristics of research areas on a science map) that aims to classify research areas in terms of continuity and cognitive linkage. Based on the Sci-GEO chart, we classified research areas into four Sci-GEO types, i.e., small island, island, peninsula, and continent, and analysed their properties, such as size and transitions across Sci-GEO types. Our analyses showed that Japanese science put more weight on continent-type research (stable and well-established research), compared with benchmarking countries, and the lack of diversity is attributable to low coverage of small island-type research (research areas with active replacement). We also demonstrated that the Sci-GEO chart well represented the funding characteristics and strategy of two major competitive funding agencies in Japan, and the chart provides policymakers and government officials with baseline information for the planning of science and technology policy.

Appendix: Details about mapping algorithm

We developed a parallel-mapping technique that can simultaneously generate maps of two different but successive periods. In this technique, a research area experiences attractive and repulsive forces that are caused by interactions with other research areas at the same and different time period, simultaneously.

Two types of attractive forces are considered. The first is the attractive forces among research areas in the same time period. The forces act on a pair of research areas connected by a co-citation linkage. An attractive force between a pair of research areas is calculated by the following relation:

$$C_{ij} \left( {{\text{Year}}\,{\text{A}}.i;\,{\text{Year}}\,{\text{A}}.j} \right) \times r\left( {{\text{Year}}\,{\text{A}}.i;\,{\text{Year}}\,{\text{A}}.j} \right)$$

(1)

$${\text{C}}_{ij} \left( {{\text{Year}}\,{\text{A}}.i;\,{\text{Year}}\,{\text{A}}.j} \right) = N\left( {{\text{Year}}\,{\text{A}}.i;\,{\text{Year}}\,{\text{A}}.j} \right) \bigg{/}\sqrt {N\left( {{\text{Year}}\,{\text{A}}.i} \right) \times {\text{N}}\left( {{\text{Year}}\,{\text{A}}.j} \right)}$$

(2)

where $N\left( {{\text{Year}}\,{\text{A}}.i;\,{\text{Year}}\,{\text{A}}.j} \right)$ represents the number of common citing papers between a research area i and a research area j in year A; $N\left( {{\text{Year}}\,{\text{A}}.i} \right)$ is the number of citing papers in a research area i in year A; and N(Year A.j) is the number of citing papers in a research area j in year A; and r(Year A.i; Year A.j) represents a physical distance between a research area i and a research area j in year A on the map.

The second is the attractive forces between research areas of two successive time periods. The forces act on a pair of research areas that have common core papers and are calculated by the following relation:

$${\text{O}}_{ij} \left( {{\text{Year}}\,{\text{A}}.i;\,{\text{Year}}\,{\text{B}}.j} \right) \times {r}\left( {{\text{Year}}\,{\text{A}}.i;\,{\text{Year}}\,{\text{B}}.j} \right)$$

(3)

$${\text{O}}_{ij} \left( {{\text{Year}}\,{\text{A}}.i;\,{\text{Year}}\,{\text{B}}.j} \right) = N\left( {{\text{Year}}\,{\text{A}}.i;\,{\text{Year}}\,{\text{B}}.j} \right)/\sqrt {N\left( {{\text{Year}}\,{\text{A}}.i} \right) \times N\left( {{\text{Year}}\,{\text{B}}.j} \right)}$$

(4)

where $N\left( {{\text{Year}}\,{\text{A}}.i;\,{\text{Year}}\,{\text{B}}.j} \right)$ represents the number of common core papers between a research area i in year A and a research area j in year B; $N\left( {{\text{Year}}\,{\text{A}}.i} \right)$ is the number of core papers in a research area i in year A; and $N\left( {{\text{Year}}\,{\text{B}}.j} \right)$ is the number of core papers in a research area j in year B; and $r\left( {{\text{Year}}\,{\text{A}}.i;\,{\text{Year}}\,{\text{B}}.j} \right)$ represents a physical distance between a research area i in year A and a research area j in year B on the map. The second attractive forces are regarded as restraints, which act as glue connecting research areas of two different periods.

We also introduced repulsive forces act on all pairs of research areas in the same period and they are calculated by β/r ², where β is a parameter to control the repulsive forces. This force is a necessity to avoid the collapse of the map. A net force for a certain research area in year A is calculated by taking a sum of repulsive and attractive forces for all pairs of research areas as the following:

$$\begin{aligned} \left( {\begin{array}{*{20}c} {{\text{F}}\left( {x_{{{\text{Year}}\,{\text{A}}.i}} } \right)} \\ {{\text{F}}\left( {y_{{{\text{Year}}\,{\text{A}}.i}} } \right)} \\ \end{array} } \right) = & - \mathop \sum \limits_{j} {\text{C}}_{ij} \left( {{\text{Year}}\,{\text{A}}.i;\,{\text{Year}}\,{\text{A}}.j} \right) \times r\left( {{\text{Year}}\,{\text{A}}.i;\,{\text{Year}}\,{\text{A}}.j} \right) \times \left( {\begin{array}{*{20}c} {\cos \;\uptheta_{ij} } \\ {\sin \;\uptheta_{ij} } \\ \end{array} } \right) \\ - \mathop \sum \limits_{l} {\text{O}}_{il} \left( {{\text{Year}}\,{\text{A}}.i;\,{\text{Year}}\,{\text{B}}.l} \right) \times {r}\left( {{\text{Year}}\,{\text{A}}.i;\,{\text{Year}}\,{\text{B}}.l} \right) \times \left( {\begin{array}{*{20}c} {\cos \;\uptheta_{il} } \\ {\sin \;\uptheta_{il} } \\ \end{array} } \right) \\ + \mathop \sum \limits_{k} \frac{\upbeta }{{r^{2} \left( {{\text{Year}}\,{\text{A}}.i;\,{\text{Year}}\,{\text{A}}.k} \right)}} \times \left( {\begin{array}{*{20}c} {\cos \;\uptheta_{ik} } \\ {\sin \;\uptheta_{ik} } \\ \end{array} } \right) -\upmu \times \left( {\begin{array}{*{20}c} {{\text{v}}_{x} \left( {x_{{{\text{Year}}\,{\text{A}}.i}} } \right)} \\ {{\text{v}}_{y} \left( {y_{{{\text{Year}}\,{\text{A}}.i}} } \right)} \\ \end{array} } \right) \\ \end{aligned}$$

where the first summation is taken over all research areas in year A connected by a co-citation linkage with a research area i in year A; the second summation is taken over all research areas in year B that have common core papers with a research area i in year A; the third summation is taken over all research areas in year A; and the fourth term is a friction term which is proportional to the velocity of a research area. Research areas with similar topics tend to locate closely according to the definition of the attractive force.

In the calculation, all research areas in Science Maps 2004 and 2006 are randomly put in different two-dimensional planes, one for Science Map 2004 and the other for Science map 2006. Then, they are moved incrementally in the direction of net force until a stable configuration is achieved. Positions of research areas are calculated by the Velocity Verlet algorithm (Swope et al. 1982). In the present analysis, the optimisation is continued until the net forces decrease less than a threshold. Several trials were conducted until the most stable results were obtained.

If the attractive forces between 2004 and 2006 maps are set to zero, there is no connection between the two maps, and they are independent. The mapping relies on attractive and repulsive forces among research areas, so only the relative locations of research areas are important. The maps for 2002, 2008, 2010, and 2012 are generated by the parallel mapping of the maps for 2004, 2006, 2008, and 2010, respectively.