Aiming at the threat assessment challenges arising from the high-dynamic, strongly adversarial characteristics of informationized air combat, as well as the limitations of traditional methods in dynamically adapting weights for threat assessment in large-scale air combat scenarios, this paper proposes a multi-level, multi-dimensional dynamic aerial threat assessment method. First, a threat assessment indicator system integrating three categories of factors—target static attributes, target maneuverability, and battlefield events—is constructed, and refined quantification modeling for each factor is completed. On this basis, an improved evaluation algorithm combining Improved Glowworm Swarm Optimization (IGSO) and Graph Neural Network (GNN) is proposed. The algorithm leverages the powerful global optimization capability of IGSO to dynamically optimize the weights and thresholds of GNN, which effectively enhances the convergence speed and prediction accuracy. The results show that the improved IGSO-GNN algorithm yields more reasonable assessment outcomes, making it better suited to the high-dynamic nature of modern battlefields, providing certain reference significance for threat assessment.