Diffusion processes

Social Networks support socio-semantic dynamics such as diffusion processes. Information circulation in knowledge networks among individuals can be traced by studying diffusion paths of semantic entity. Studying real-world diffusion dynamics can also be important to unveil influence dynamics between agents, find opinion leaders, etc.

Contrary to traditionnal epidemiologic models which assume that populations are homogeneous or  geometrically organized, real-world diffusion processes in knowledge networks highly depend on the topology of the underlying social network. For a asynchronous diffusion process, we observe that highly clustered networks slow the spreading, while the distribution of degrees of nodes does not seem to have significant impact on diffusion speed.

Influence dynamics between groups of sources can be reconstructed thanks to machine learning analysis which track systematic intertemporal correlations  between contents published by sources. For example it allows to build an influence map between groups of blogs, answering questions such as: which group is imposing its agenda over other groups?

Real-world diffusion of URLs form cascades of citations in the blogosphere. There are strong correlations between cascades structure and the topology of the blog network. Studying both various spreading dynamics and network topology allows to define individual or meso-level characteristics that make a good spreader.

Data is a key point to describe satisfactorily real-world process. It may be necessary to “build” an experimental set-up which allows to gather complex data. Happyflu is such a real-world experiment. It proposed bloggers to copy an applet on their page allowing to track its diffusion over about 400 websites.

Comments are closed.