To refine the effectiveness and accuracy of high-rise building construction safety assessments, this study introduces an innovative method that incorporates a graph neural network. Firstly, a comprehensive safety assessment framework is devised, encompassing three distinct levels: target (A), criterion (B), and index (C). Secondly, the weighting of safety evaluation metrics is meticulously determined, allowing for dynamic adjustments to both fixed and variable weights. Utilizing the powerful capabilities of the graph neural network, we extract pertinent characteristics of high-rise construction projects. Building upon this foundation, we perform dynamic safety evaluations specific to high-rise buildings, ultimately generating safety assessment outputs. Rigorous testing reveals that our method surpasses conventional approaches in terms of accuracy, recall, precision, and F1 score, underscoring its remarkable efficacy and superiority.