In this study, we tackle the geolocalization problem of identifying the location where a photograph was taken. Conventional methods often involve multiclass classification using area meshes, which means that the correlation between the photographs and the area classes is weak, limiting the performance of the extracted features. In this study, we propose a method to divide areas by constructing a proximity graph by connecting photography points of interest (POIs) and extracting communities. This strengthens the correlation between the image feature and the area class, making it possible to extract meaningful features. Through evaluation experiments using real data, we compare the differences between proximity graphs, and the differences between the proposed method and existing area segmentation methods such as administrative divisions.