Information Diffusion Prediction with Personalized Graph Neural Networks.

Online social networks are crowded with massive information, which is more likely to spread rapidly on a large scale. Therefore, understanding and predicting information diffusion on social networks will be much helpful to improve the performance of marketing and control the dissemination of misinformation. Recently, the deep learning techniques have enhanced the methods for diffusion prediction and provide a new way to model the diffusion process in time and space. However, these models introduce the temporal and structural factors affecting the diffusion with two sequential or parallel steps separately. Moreover, they neglect the whole influence of the diffusion cascade from a global view. Hence, we propose a novel method for diffusion prediction with personalized graph neural networks, namely infGNN, to model the interactions between structural and temporal factors. Furthermore, we integrate local and global influence of diffusion cascade for prediction. Experiments results on three datasets show the superiority of the proposed model.