Session on Social Computing - IEEE SocialCom'09

I'm now in the session on Social Computing, one of the sessions in IEEE SocialCom conference. The first talk that I attended was about Guangxi on the Chinese Web where the presenter showed work that they did in doing an empirical study of Guangxi web sites using PageRank, compared to the general web and structures that come about from the Guangxi web sites by modelling Guangxi and identifying Guangxi links. The question that comes to mind, is how do the Guangxi web social networks compare with regular social networks, and can this help explain why Facebook and Twitter are not popular with Chinese users?

Another interesting paper was on Structure of Neighborhoods in a Large Social Network using a dataset obtained from Orange mobile phone users. They used "characteristic patterns" to identify neighbourhoods. The first talk in the second session on Social Computing is on Deriving Expertise Profiles from Tags. Their assumptions are that the set of tags defines a resource, and that these tags are correlated with skills. They performed a study with Dogear and IBMr, internal IBM social computing systems. They also create a scoring model to correlate tags with skills, and did find out that the tags do represent the skills. It would be interesting to see how finding similar users and experts could be used for tag recommendations. The third paper is on Probabilistic Generative Models of the Social Annotation Process addressing the challenge of uncovering hidden structure in social annotations, can we discover communities of users and categories of related tags. Their inspiration is on text-based topic models and their solution is Community-Based Probabilistic Social Annotation (PSA), instead of modelling tags, they also model users. Users belong to hidden communities so how can we find these communities, also related to my paper that I will present tomorrow. They recover communities and categories based on the Gibbs Sampler and used Delicious for their experiments. The PSA is better at predicting unsessn data, and they also did a user study to determine if the tags were correctly specified in the categories.

The next paper is on Detecting Communities from Bipartite Networks Based on Bipartite Modularities.