Against the backdrop of sustained economic growth, certain regions and consumption sectors in China have exhibited downgrading trends, yet existing research lacks exploration of the characteristics of different consumption forms and their spatial drivers. This study analyzes the spatial differentiation of online and offline consumption downgrading in 289 Chinese cities from 2018 to 2023, and explores the mechanisms of multidimensional socioeconomic factors. The research constructs a Geographic Convolutional Neural Network Weighted Regression (GCNNWR) model, integrating the nonlinear feature learning of convolutional neural networks with the spatial heterogeneity capturing of geographically weighted regression. The entropy method is employed to weight online and offline consumption indicators, and socioeconomic variables are integrated to build an explanatory framework. The research found: (1) Urban consumption downgrading in China presented significant spatial differentiation patterns, with offline consumption downgrading being more concentrated and intensive in high economic agglomeration areas such as Beijing-Tianjin-Hebei and the Yangtze River Delta, while online consumption downgrading exhibited stronger spatial diffusion effects; (2) Consumption downgrading displayed an “inverse gradient” distribution phenomenon, where economically developed regions actually exceeded underdeveloped regions in their degree of consumption downgrading; (3) Economic agglomeration level had the strongest explanatory power for offline consumption downgrading, while population density ranked second in its impact on online consumption downgrading, indicating fundamental differences in the spatial agglomeration effect mechanisms of different consumption modes; (4) Higher education levels showed positive correlations with consumption downgrading, and the consumption-promoting effects of industrial structure optimization exhibited obvious regional threshold characteristics. This study provides evidence for understanding urban consumption transformation. • Offline downgrading clusters in agglomeration areas; online shows spatial diffusion. • Developed regions show greater downgrading, revealing an "inverse gradient" pattern. • Economic agglomeration drives offline downgrading; population density drives online. • Education level links to downgrading; industrial optimization has threshold effects. • GCNNWR model combines CNN learning with GWR spatial analysis for behavior modeling.
Rui et al. (Tue,) studied this question.