This paper proposes a dynamic threat assessment method for aerial targets by integrating graph embedding features and a Gated Recurrent Unit (GRU) network. To address the limitation of traditional threat assessment methods in overlooking collaborative relationships among multiple targets, we leverage graph embedding technology to effectively represent both the attribute features and inter-dependencies of aerial targets. A GRU network is then employed for temporal sequence analysis, enhancing the accuracy and real-time performance of intent recognition for aerial target behaviors. Building on the intent recognition results, a threat assessment indicator system is constructed using the Analytic Hierarchy Process (AHP), enabling quantitative evaluation of threat levels. Experimental results demonstrate that the graph embedding technique significantly improves the effectiveness of target intent recognition. The introduction of intent recognition results markedly reduces threat ranking errors and enhances the accuracy of aerial target threat assessment, providing robust support for subsequent task allocation decision-making.