Relation-enhanced Diffusion Model with LightGCN for Information Diffusion Prediction

Abstract Modeling and predicting the information diffusion process on social platforms is a critical task in many real-world applications. Recent studies generally model graph using neural networks to capture implicit dependencies among users. However, existing construct way which cannot fully describe global of users due their narrow definition user relationship. Meanwhile, these methods are not suitable for network scenario has scarce node attributes. Therefore, we propose novel construction method can enhance relations adopt simplified convolutional operation prediction scenario. The learned embedding our effectively preserve microscopic structure high-order proximities between lies both graph. Experimental results four datasets show that proposed superior most advanced methods.