Bayesian Influence Centrality: Identify the Right Individuals for Information Diffusion
The influence maximization literature has long considered maximizing information diffusion coverage as the main objective. Banerjee (2014) and Abebe (2018) focused on two important factors respectively, diffusion probability and evaluation on the information/product, which have been overlooked by most studies in the field. Building upon these two studies, we propose a Bayesian Influence Centrality (BIC) which measures the expected relative influence people exerted in the social network based on their network positions, the likelihood of spreading the information, as well as their evaluations of the product. We show that BIC nests degree centrality, Katz centrality, eigenvector centrality and diffusion centrality. Meanwhile, we propose a measure to select the connected components to target in social networks using the Laplacian quadratic form to study how likely a campaign will be successful within the component depending on the three main features mentioned above. Moroever, modeling diffusion probability and evaluations using Normal distribution, we provide an iterative analytical solution for the mean and variance of BIC. We simulate an information diffusion process on multiple types of real-world social networks and show how BIC out-performs existing centrality measures. The proposed BIC measure has important implications for and applications in marketing and political campaigns, as well as large-scale behavioral changes.
Author: Yan Leng