Structural virality estimation and maximization in diffusion networks

Abstract

Social media usage is one of the most popular online activities and people shares millions of message in a short time; however this information rarely goes viral. The diffusion process begins with an initial set of source nodes and continues with other nodes. In addition, the viral cascade is triggered when the number of infected nodes exceeds a specific threshold. Then, we find an initial set of source nodes that maximizes the number of infected nodes given the source nodes. This study aims to answer the following questions: how does a spread like a viral cascade propagate in a network? Do the structural properties of the propagation pattern play an important role in virality? If so, can we shape the propagation pattern to maximize the final influence? In this regard, we introduce two different extreme cascades: first, the broadcast process starts when a famous person infects a large number of his followers. Second, the person-to-person process propagates information to a large number of people and each node infects a small number of nodes. In detail, we characterize the structural properties of each diffusion pattern and, specifically, we are using structural virality measure to evaluate how much a diffusion pattern is close to either broadcast or viral cascade. Finally, we develop an efficient viral structure estimation and maximization method, ViSEM, that maximizes the structural virality with two surrogate functions for different diffusion models. We show both empirically and theoretically that the proposed method approximates a solution with a provable guarantee.