Abstract: With the increasing popularity and wide application of social networks, information diffusion prediction has gradually become a hot research topic in the field of social network analysis. Most previous studies either only use the information diffusion sequence or only use the social network between users to make prediction, failing to effectively model the complexity of the information diffusion process. In addition, recurrent neural network (RNN) and its variants, which are commonly used in information diffusion prediction, are difficult to capture the correlation between information effectively. To address the above problems, we propose a novel social network information diffusion prediction model called STT based on spatial-temporal Transformer. First, we construct a heterogeneous graph composed of a social network graph and a dynamic diffusion graph, and use graph convolutional network (GCN) to learn the users’ structural features. Then, the users’ temporal features and structural features are put into the Transformer for fusion to obtain users’ spatial-temporal features. In order to effectively fuse the users’ temporal features and structural features, a novel residual fusion method is proposed to replace the original residual connection in Transformer. Finally, the Transformer is used for information diffusion prediction. Extensive experiments on real datasets demonstrate the effectiveness of our proposed model STT.