XAI in geographic analysis of innovation: Evaluating proximity factors in the innovation networks of Chinese technology companies through web-based data

Abstract

This research explores the nonlinear interactions among multidimensional proximities, including geographical, cognitive, organizational, institutional, social, and technological aspects, and their impact on innovation within networks of over three million technology firms in China. Utilizing an innovative combination of web-based hyperlink and textual data analysis, supplemented by patent information, we delve into how these proximity dimensions influence corporate innovation capabilities. Our methodology integrates text-based deep learning techniques and employs the XGBoost model along with the SHapley Additive exPlanations (SHAP) algorithm and partial dependence plots to uncover the nuanced effects of proximity on innovation. The findings reveal that while geographical distance often correlates with larger cognitive and organizational proximities, underdeveloped regions exhibit stronger technological, institutional, and social proximities compared to their developed counterparts. The study further identifies social structure and technological differences as pivotal factors impacting collaborative innovation, with both positive and negative effects fluctuating alongside changes in proximity dimensions. Notably, we uncover that geographical proximity has a pronounced boundary effect on innovation, highlighting the critical role of spatial considerations in the digital age of innovation networks. This research contributes to the understanding of urban innovation dynamics and offers valuable insights for policymakers and urban planners aiming to foster innovation ecosystems.