Emering problems in influence propagation and maximization

Abstract

In the last two decades we witnessed the advent and the rapid growth of online social networks (OSNs). The impact of their pervasive diffusion on everyday life has been dramatic. In fact, social networks changed the way we interact with each other, the way we access information and the way companies engage with their audience or customers. A major consequence of the broad adoption and diffusion of social networks is the availability of an unprecedented amount of user data, which enables the opportunity for social and network scientists to investigate and observe many facets of human behaviors. Arguably, one of the most interesting facet is related to the notion of social influence. Following this observation, this research project is mainly centered around the concept of social influence, specifically its propagation and maximization. Therefore, the goal of this thesis is twofold. To begin with, we investigate the complexity of influence propagation in real-world contexts. This leads to the definition of a novel class of diffusion models. Such models represent an attempt to unify, under a well-defined framework, all the aspects that contribute to the inherent complexity of any influence propagation phenomena. Afterwards, we devote our attention to the influence maximization problem. To this purpose, we first provide a detailed characterization of social influence from a topological perspective. Specifically, we want to understand if and to what extent being a good spreader depends on being located into strategic regions of a network. Finally, we focus on the application of the influence maximization problem. In particular, we address a variant of the original problem, which is especially suitable for viral marketing scenarios. To this end, we propose two different diversity-sensitive targeted influence maximization problems. Both proposals share a common intent, which is assessing the benefit of embedding a notion of diversity into the process of the seeds identification. Nonetheless, diversity is considered from two different perspectives: (i) as a function of the topological properties of the nodes; (ii) as a function of some categorical data available on the node level.