As an important and fundamental national resource, government data resources are of strategic importance in promoting economic growth and social development. This paper discusses the utilisation model of open government data in order to build an analysis framework for an open government data ecosystem. Starting from the government-led internal management utilisation model, the open government traffic accident data is selected, and a prediction model based on the scale-reduced attention mechanism and graph convolutional network is proposed by using deep learning algorithms. Then build a research model of open government data usage, combined with regression analysis to study the influence mechanism of open government data usage. Through experimental analyses, the SAGCN model in this paper demonstrates better traffic accident prediction; its MAE (0.082), MSE (0.038), and MRE (0.808) are smaller than that of the comparison model, and it has better prediction ability under busy traffic and weather anomalies. In addition, the completeness, quality, and risk attitude of deep-level facilities positively affect shallow-level perceived behavioral control, which in turn affects the utilization of open government data.