Learning Influence Probabilities and Modelling Influence Diffusion in Twitter

Zizhu Zhang, Weiliang Zhao, Jian Yang, Cécile Paris, Surya Nepal

Abstract: Influence diffusion has been widely studied in social networks for applications such as service promotion and marketing. There are two challenging issues here: (1) how we measure people’s influence on others; (2) how we predict whom would be influenced by a particular person and when people would be influenced. Existing works have not captured the temporal and structural characteristics of influence diffusion in Twitter. In this paper, we firstly develop a model to learn influence probabilities between users in Twitter from their action history; secondly, we introduce diffusion models that are used to predict how information is propagated in Twitter. Experiment results show that our proposed models outperform existing models in terms of the balanced precision and recall.