Information Diffusion Predictive Model Using Radiation Transfer

Abstract

The impact of online social networks on information exchange between humans has revealed the need to study the mechanisms of information diffusion. Multiple prior works have considered empirical studies and introduced new diffusion models to understand the dynamics of the diffusion process. However, the complexity of network structures and user interactions make it challenging to model the diffusion mechanisms of online social networks and to accurately predict diffusion. In this paper, we propose an information diffusion prediction model based on a physical radiation energy transfer mechanism. The aim of this model is to predict the diffusion graph of a certain contagion throughout an interest-based community. This non-parametric model can accommodate the dynamicity of online social networks because it can receive different input diffusion parameters at different diffusion contagions. With our RADiation DIFFusion (RADDIFF) model, we precisely capture the information diffusion process from both temporal and spatial dimensions and measure the level of influence initiated by certain influencers for each diffusion process. To our knowledge, this model is the first in this domain that exploits the prediction of information networks based on a physical radiation mechanism. We conduct an extensive analysis using an experiment that includes two well-known prediction diffusion models, the linear influence model (LIM) and NETINF. The results show that RADDIFF effectively outperforms both the LIM and NETINF in terms of accuracy and the quality of forecast.