Abstract: Electroencephalogram (EEG) emotion recognition holds significant applications value in human-computer interaction and diagnosis and treatment of emotional disorders. However, existing methods for characterizing brain emotional activities have limitations in fully exploring the dynamic and directed spatial-temporal features within EEG. To address the insufficient modeling of dynamic topological relationships among brain regions in EEG-based emotion recognition studies, this research proposes a Dynamic Directed Spatial-Temporal Graph Network. This innovative approach constructs dynamic directed spatial-temporal graphs to capture the dynamic changes in topological connections between EEG channels during emotion induction processes, thereby fully exploiting spatial-temporal features. Experiments evaluations on the SEED and SEED-IV public datasets demonstrate that the proposed method outperforming state-of-the-art models, indicating its significant advantages in EEG emotion recognition tasks.