Session on Networked CommunitiesThe

The first paper in this session on Networked Communities is Modularity in Heterogeneous Networks by Tsuyoshi Murata. This is very important because Newman-Girvan modularity is usually only performed on homogeneous networks and bi-partite graphs. The authors propose a tripartite modularity and optimization of their measure. Their future work is looking into how to apply this and validate this with a real life dataset like for example Delicious.

The second paper is Link Prediction Applied to an Open Large-Scale Online Social Network by Dan Corlette and Frank Shipman. Can we build a model of large online social networks? They view the network as a graph using link topology. For them, an online social network is user generated content and list of friends. Their model is used for prediction. They use LiveJournal as their dataset and used a naiive Bayes classifier for training the dataset. Their main results is that the difficulty of predicting new friendships grows the longer users have been members of the network. Currently, they are building the model, their future work involves using user interests and centrality measures for improving link prediction.

The third paper is Community-Based Ranking of the Social Web by Said Kashoob, James Caverlee. and Krishna Kamath. Their research questions on do user-based communities manifest themselves in social bookmarking systems and how to model them? Their hypothesis is community-based tagging and the problem is how to uncover underlying communities given only observed tags and users. They create their own model called the CTAG model that emphasize on user role. They use the Gibbs Sampler for their model. For each community, they get a distribution over users and distribution over tags. They use experimental likelihood to test their CTAG model.

The last paper is Social Networks and Interest Similarity: The Case of CiteULike by Danielle H. Lee and Peter Brusilovsky. They use unilateral relationships as edges for the social network like "following" on Twitter, "watching" on CiteULike, this is a one-way relationship and does not require mutual agreement about being in the relationship. They used information similarity in order to find interest similarity, and with application to CiteuLike, users undershared items.