Extracting and ranking viral communities using seeds and content similarity

We study the community extraction problem within the context of networks of blogs and forums. When starting from a small set of known seed nodes, we argue that the use of content information (beyond explicit link information) plays an essential role in the identification of the relevant community. Our approach lends itself to a new and insightful ranking scheme for members of the extracted community and an efficient algorithm for inflating/deflating the extracted community. Using a considerably large commercial data set of blog and forum sites, we provide experimental evidence to demonstrate the utility, efficiency, and stability of our methods.