Shared mobility redefines urban transportation, offering economic and environmental benefits by reducing pollution and urban congestion. However, in the post-pandemic era, the shared mobility sector is grappling with a crisis of trust, particularly concerning passenger hesistancy towards shared transportation options. To address these problems, in this paper we take social network into consideration and propose a novel carpooling matching framework based on graph neural network and reinforcement learning,increasing the carpooling rate to 48% and reducing the average delay time to 6.1 minutes and average detour distance to 2.8km. Furthermore, we introduce an innovative metric, termed 'tolerance' for mobility scheduling models to effectively quantify users' sensitivity to social distancing. We conduct a sensitivity analysis to demonstrate that our model offers a viable approach to amplify the benefits, delivering resilient strategies for the advancement and proliferation of shared mobility incentives.