In order to explore how the digital economy reshaped the innovation ecosystem of Shuangcheng Economic Circle in Chengdu-Chongqing region, this paper proposed an artificial intelligent-driven construction framework that integrated multi-source innovation factor clustering, spatio-temporal graph neural network and dynamic feedback update. Firstly, the framework divided enterprises, research institutions, platforms, funds and policy resources into collaborative units from heterogeneous regional data. Then, cross-regional knowledge flows, industrial linkages, and innovation interactions are modeled through spatio-temporal graph representations to infer ecosystem association strength. The feedback update module is further introduced to correct the structural bias and stabilize the evolution estimation. Experiments were carried out on 42,318 records collected from 38 districts and counties. The results show that the proposed model achieves 93.4% construction accuracy and 91.7% collaborative recall rate, and the overall performance is stable better than that of comparison methods and similar models. The verification results provide a computable analysis basis for the regional innovation governance, resource allocation and smart policy adjustment of the twin-city innovation network under the background of digital transformation.
