Model-based learning of information diffusion in social media networks

Abstract

Social networks have become widely used platforms for their users to share information. Learning the information diffusion process is essential for successful applications of viral marketing and cyber security in social media networks. This paper proposes two learning models that are aimed at learning person-to-person influence in information diffusion from historical cascades based on the threshold propagation model. The first model is based on the linear threshold propagation model. In addition, by considering multi-step information propagation in one time period, this paper proposes a learning model for multi-step diffusion influence between pairs of users based on the idea of random walk. Mixed integer programs (MIP) have been used to learn these models by minimizing the prediction errors, where decision variables are estimations of the diffusion influence between pairs of users. For large-scale networks, this paper develops approximate methods for those learning models by using artificial neural networks to learn the pairwise influence. Extensive computational experiments using both synthetic data and real data have been conducted to demonstrate the effectiveness of the proposed models and methods.