Cascade of Information in Networks
The propagation of information and new ideas has long been a fundamental question in the social sciences. Propagation may be driven by exogenous causes, when people are informed by an external source, e.g. television, but also by endogenous mechanisms, when a few early adopters may influence their friends, who may in turn influence their own friends and possibly lead to a cascade of influence. This self-organizing process, which reminds of the dynamics of an epidemic, is usually called the word-of-mouth phenomenon. It has attracted more and more attention in the last few years due to the emergence of the internet and of online social networks, which have led to more decentralized media of communication. A typical example is the blogosphere, where blogs are written and read by web users and where debates/discussions may take place among the bloggers. As of today, the blogosphere is extremely influential in the adoption or rejection of products but also in politics, as more and more citizens voice their opinions and mobilize community efforts around their candidates. From a practical point of view, the emergence of these participative media has changed the way elections take place, by allowing politicians to reach new audiences, raise money and communicate to voters, and also to open new ways to promote commercial products via recommendation networks or viral marketing methods. It is therefore primordial to better understand how such information cascades take place in social networks.
A good description of the word-of-mouth phenomenon requires two elements: a model of propagation and a network structure. The model of propagation defines the way information (e.g. a marketing campaign for a specific product, an information) flows between acquaintances, and is usually based on models for epidemic spreading (with susceptible/infectious/removed individuals). Second, this viral process has to be applied on a realistic social network, where each node defines a member of the society and edges are drawn between acquaintances. For a long time the design of these social networks was purely theoretical and real social networks were generally limited in size, but the advent of the Internet and of cheap computer power now allows to study social networks composed of millions of individuals and to characterize the statistical properties of their topology. An important challenge is therefore to understand how the topology of the social network affects the propagation of information but also to find statistical indicators for the most influential nodes in the network, and models describing the formation of the social network itself.
The panel will consist of Paul Bourgine (CREA), Jeffrey Johnson (Open University), Renaud Lambiotte (Imperial College) and Jure Leskovec (Carnegie Mellon) and will investigate questions such as
- Do social networks (or information networks) have some universal properties?
- Is it really possible to find indicators for the most influential people?
- What kind of experiments can be performed in order to validate the models?
- What has it to do with statistical physics?