Skip to main content



Gesundheit! Modeling Contagion through Facebook News Feed

This paper empirically asses the conditions under which large-scale cascades occur within a social network– specifically Facebook. It also further analyzes cascade differences in social media as opposed to cascades resulting from an isolated event. The study focuses on information diffusion of Facebook Pages– representation for businesses, bands, celerities, etc. as distinct, customized profiles. Traditionally, a number of network models use the assumption that a small number of nodes triggers a large chain-reaction or cascade (as covered in class), however in social media systems, cascades are caused by publicly visible pieces of content that are introduced from many otherwise disconnected sources. Hence, instead of information spread acting as long, branching chains of adoption, diffusion will instead be characterized by large-scale collisions of shorter chains. Information on characteristics that affect a node’s maximum diffusion chain length is valuable insight to marketers as well future research on public opinion formation.

It was found that a new characterization of global cascades may be needed in order to describe the diffusion clusters observed on Facebook: global cascades are events that begin at a large number of nodes who that initiate short chains in which each of these chains quickly collide into a large single structure. Further, it was found that a user’s demographics or Facebook usage characteristics cannot meaningfully predict that start node’s maximum diffusion chain length (in an information cascade).

Lectures in INFO 2040 have discussed information cascades using a model with the assumption that a small number of nodes triggers a large chain-reaction or cascade. This study pushes on traditional information diffusion modeling by analyzing large social network data (262,985 Facebook Pages and their associated fans). Further, there the study concluded there is not a clear way to identify initiators that are more likely to trigger a large global cascade. In INFO 2040 it was discussed that the likelihood of a global cascade varies given an initial node depends in part on the level of influence and connectedness of the node. While the results of the study are highly tied to the unique setting, it is important to consider the contrary findings in relation to common diffusion modeling assumptions.

Paper source here: https://www.aaai.org/ocs/index.php/ICWSM/09/paper/view/185

Comments

Leave a Reply

Blogging Calendar

November 2017
M T W T F S S
 12345
6789101112
13141516171819
20212223242526
27282930  

Archives