CIP: Community-based influence spread prediction for large-scale social networks

Abstract

Understanding information diffusion and influence spread remains one of the key problems in network science. Many algorithms and techniques have been proposed to maximize information diffusion and identify key opinion leaders in a given network. A central idea in most of these works is based on the “influential hypothesis” – that a small number of influencers can induce information cascades that spread to a significant portion of the population. However, works in social network analysis have shown that a critical mass of individuals who can be easily influenced (susceptible) is also crucial to drive influence cascades in a population. In this paper, we propose a novel community-based influence spread prediction (CIP) methodology to understand and predict influence spread based on individuals’ susceptibility to influence. The proposed approach leverages inherent attributes in social networks such as individual traits and community structure to help understand and predict influence spread. In addition, the CIP methodology also presents many natural avenues for parallelization and scalability which are crucial for handling real-world large-scale social networks. Preliminary comparisons with simulated influence spread using state-of-the-art influence maximization algorithms on a real-world large-scale physician network show that the CIP method can predict influence spread with a difference of less than 7%. The findings also demonstrate that the CIP approach can be used effectively to study influence spread characteristics and aid in the formulation of strategies to reach certain demographic groups or communities.