Finding and Evaluating Community Structures in Spatial Networks

Abstract

Community detection can reveal unknown spatial structures embedded in spatial networks. Current spatial community detection methods are mostly modularity-based. However, due to the lack of appropriate spatial networks serving as a benchmark, the accuracy and effectiveness of these methods have not been tested sufficiently so far. This study first introduced a spatial autoregressive and gravity model united method (SARGM) to simulate benchmark spatial networks with known regional distributions. Then, a novel spectral clustering-based spatial community detection method (SCSCD) was proposed to identify spatial communities from eight kinds of benchmark spatial networks. Comparative experiments on SCSCD and three other methods showed that SCSCD performed the best in accuracy and effectiveness. Moreover, the scale parameter and the community number setting of the SCSCD were investigated experimentally. Finally, a case study was applied to the SCSCD to demonstrate its ability to extract the internal community structure of a high-speed train network in China.