Abstract: ‍ ‍In recent years， graph neural networks have shown outstanding performance in the field of electroencephalography （EEG） emotion recognition， owing to their excellent spatial topology abilities. However， many methods only consider the connections between EEG channels and ignore the functional connections between brain regions. Graph neural networks typically rely on linear transformations to update nodes， and directly using original EEG signals as input can easily lead to the overfitting of the model. Although manually extracted features can alleviate this problem， they also limit the comprehensive extraction of information. In response to the limitations of current graph related algorithms in the field of EEG emotion recognition， this paper proposes an end-to-end EEG emotion recognition algorithm based on adaptive hierarchical graph pooling technology. First， an adaptive graph generation module that adopts a data-driven approach to automatically generate graph adjacency matrices closely related to emotion recognition tasks is designed to effectively model the spatial information of EEG signals. In addition， to more accurately extract the time-domain EEG signal features of key channels or brain regions， this study further proposes a time-domain hierarchical graph pooling module that combines the relevant features of graph nodes while identifying time-domain signal features. This module extracts more representative regional features through the layer-by-layer pooling of EEG channels and dynamically updates the functional connections between regions， thereby achieving deep feature extraction and the effective representation of EEG signals. Experiments were conducted on two widely used public EEG emotion datasets， namely， DEAP and MAHNOB-HCI， to verify the effectiveness of the proposed adaptive hierarchical graph pooling model. The results showed that the proposed adaptive hierarchical graph pooling model significantly outperformed other state-of-the-art deep learning and traditional machine learning methods in terms of performance.