Modeling Contagion Through Facebook News Feed
Citation
Eric Sun, Itamar Rosenn, Cameron A. Marlow and Thomas M. Lento. Gesundheit! Modeling Contagion through Facebook News Feed, AAAI 2009.
Online version
Summary
This paper presents novel findings about how the action "liking a Page" difuse in the Facebook social network. The authors analyzed all the Facebook Pages created between Feb 19, 2008 and Aug 19, 2008, and calculated the length of the "like chain" and the "initiator" vs "follower" demographics of a Page to model how the like-a-Page action diffuses across Facebook. They also developed a novel way of predicting the maximum length of a "like chain" using zero-inflated binomial regression.
Discussion
The authors clustered the fans of certain Facebook Page by connectivity (there's a friend path between any two fans in the same cluster), and claimed that in most clusters, 14% - 18% of the nodes (people) are chain initiators (they liked the Page by searching for it themselves). This percentage is much higher than what others have observed in other social norms. Also, The maximum length of a "like chain" is much longer than the word-of-mouth case, with 86.4% of the "like chains" have at least 4 nodes, and the longest chain being 82-node long. This may provide marketers some useful information when they decide where to put their ads next time.
What also contradicts with previous assumption is the property of decentralization. The diffusion of the like-a-Page action is not correlated with how active a user is or how many friends he/she has. The only controlling factor here is the likelihood of his/her like-a-Page action to appear on his/her friends' news feeds. Also, instead of starting from a small number of initiator nodes and spread out to their adjacent nodes, the global diffusion cascades starts with a large number of nodes who initiate short chains; each of those chains then quickly collide into a large single cluster.
Related Papers
- Watts, D. J. and Dodds, P. S. 2007. Influentials, Networks, and Public Opinion Formation. Journal of Consumer Research 34: 441-58.
- Leskovec, J. et al. 2007. Cascading Behavior in Large Blog Graphs. In SIAM International Conference on Data Mining.
Study Plan
You may first want to take a look at Binomial regression and Poisson regression.
Also, read the papers in the "Related Papers" section.
