Skip to main content



Information Cascade at Group Scale

In this paper, the authors aim at locate the most “influential groups” in stead of individuals in a network to trigger network diffusion. They firstly propose a fine-grained model of information diffusion for the group-based problem, and then show that the process is submodular and present an algorithm to determine the influential groups under this model. After that, they propose a coarse-grained model that inspects the network at group level significantly speeding up calculations for large networks They also show that the diffusion function they design here is submodular in general case, and propose an approximation algorithm for this coarse-grained model, and finally by conducting experiments on real datasets. At the end of the story, they demonstrate that seeding members of selected groups to be the first adopters can broaden diffusion. Moreover, they can identify these influential groups much faster, delivering a practical solution to this problem.

I personally quite agree the opinion that we should focus on instigate groups rather than individuals in some real cases. Those cases abound, e.g., billboards, TV commercials and newspaper ads are utilized extensively to boost the popularity and raise awareness. Groups and associations are natural targets of initial convincing attempts in may real-life scenarios. And the models in the paper are also reasonably constructed, such as FGD and CGD. Under rational assumptions, they show that they can achieve wider diffusion and faster speeds by focusing on groups. They also find out group algorithms run much faster than individual algorithm.

Although the CGD model aggregates the information about individuals and ignores many details, it result in a final influence comparable to the FGD model. In fact, in the individual diffusion model as well as the fine-grained group diffusion model, they identify k entities that are highly influential for different values of nodes’ thresholds theta, which is an interesting point to dig into. So the authors provide several future research directions based on this paper’s results.

http://www.cs.toronto.edu/~milad/paper/fp1006-eftekhar.pdf

Comments

Leave a Reply

Blogging Calendar

November 2015
M T W T F S S
 1
2345678
9101112131415
16171819202122
23242526272829
30  

Archives