Visualizing large-scale structure of a million-firms economic network Yuji Fujita Turnstone Research Institute Yoshi Fujiwara University of Hyogo Wataru Souma Nihon University
Overview Graph drawing method proposal Application to the Real Economic Network of Japan – 1.2 mils of companies – 5.2 mils of links ● Supplier/customer ● Shareholding – 500 bankruptcies since the Earthquake of 2011
Graph(network) drawing ● No naiive approach for large scale data 1: Graph structure based optimize 2: Embedded space based optimize ● In this study: Hybrid of 1 and 2. Graph structure gives initial configuration, then force-directed method applied.
Graph visualization criteria ● To recognize feature ● To identify
Hybrid for quality and efficiency ● MDS gives initial coordinate – PMDS optimizes computation cost ● Force-directed method improves quality – Barnes-Hut hierarchical algorithm reduces n**2 to n*log_n Identify Distinguish MDS Good So-so Force NG Good
Overall view in motion
Industrial sectors Manufacturer Construction Infrastructure Wholesale
Conclusion ● Hybrid algorithm of MDS and force-directed ● Real economic network of Japan visualized – Modular structure of industrial sectors – Bankruptcy information visualized
Acknowledgements ● Grant-in-Aid for Scientific Research (KAKENHI) Grant Numbers 25282094 by JSPS. ● Data is provided by the Research Institute of Economy, Trade and Industry (RIETI) in connection to the project ``Price Network and Dynamics of Small and Medium Enterprises''