SkipCas: Information Diffusion Prediction Model Based on Skip-Gram

Abstract

The development of social network platforms such as Twitter and Weibo has accelerated the generation and transmission of information. Predicting the growth size of the information cascade is widely used in the fields of preventing rumor spread, viral marketing, recommendation system and so on. However, most of the existing methods either cannot fully capture the structural representation of the cascade graph, or cannot effectively utilize the dynamic changes of information diffusion, which often leads to poor prediction results. Therefore, in this paper, we propose a novel deep learning model called SkipCas to predict the growth size of the information cascade. First, we use the diffusion path and time effect at each diffusion time in the cascade graph to obtain the dynamic process of the information diffusion. Second, we put the sequence of biased random walk sampling into the skip-gram model to obtain the structural representation of the cascade graph. Finally, we combine the dynamic diffusion process and the structural representation to predict the growth size of the information cascade. Extensive experiments on two real datasets show that our model SkipCas significantly improves the prediction accuracy compared with the state-of-the-art models.