This paper proposes a dynamic semantic-based graph convolutional network (DS-GCN) for skeleton-based human action recognition. The proposed method encodes joint and edge types in an implicit way within the skeleton topology, enabling more accurate representation of semantic information. Two semantic modules, the joints type-aware adaptive topology and the edge type-aware adaptive topology, are introduced to capture the semantic properties of actions. These modules are combined with temporal convolution to form a powerful framework for skeleton-based action recognition. Extensive experiments on the NTU-RGB+D and Kinetics-400 datasets show that the proposed DS-GCN outperforms state-of-the-art methods in terms of recognition accuracy. The DS-GCN is able to generalize well across different backbone structures and effectively captures the dynamic nature of human actions. The method is designed to adaptively learn the skeleton graph structure based on the semantic information of joints and edges, leading to improved performance in action recognition tasks. The code for the proposed method is available at https://github.com/davelailai/DS-GCN.This paper proposes a dynamic semantic-based graph convolutional network (DS-GCN) for skeleton-based human action recognition. The proposed method encodes joint and edge types in an implicit way within the skeleton topology, enabling more accurate representation of semantic information. Two semantic modules, the joints type-aware adaptive topology and the edge type-aware adaptive topology, are introduced to capture the semantic properties of actions. These modules are combined with temporal convolution to form a powerful framework for skeleton-based action recognition. Extensive experiments on the NTU-RGB+D and Kinetics-400 datasets show that the proposed DS-GCN outperforms state-of-the-art methods in terms of recognition accuracy. The DS-GCN is able to generalize well across different backbone structures and effectively captures the dynamic nature of human actions. The method is designed to adaptively learn the skeleton graph structure based on the semantic information of joints and edges, leading to improved performance in action recognition tasks. The code for the proposed method is available at https://github.com/davelailai/DS-GCN.