Communities in Social Media: Reflections on Semantics, Intention and Influence

Communities are central to online social media systems and detecting their structure and membership is critical for many applications. A community in real world is represented in a graph as a set of nodes that are more closely related to one another than the rest of the network. In social media, a community could be a set of blogs that are topically related, a group of friends connected via Live Spaces or even a set of users who share similar tags in their social bookmarks. Graph structure has commonly been used to detect communities. However, we can go beyond that by utilizing the special properties and meta-data available in social media to identify such communities. For instance, due to the sparsity and long tail structure of social graphs it is possible to efficiently estimate communities by sampling only a small portion of the entire graph. Another useful property of social media datasets is the availability of tags, which provide free meta-data. Community detection can benefit from not just how nodes link to each other, but also what tags they use. Grouping blogs or feeds via tags can help describe the topics that relate the set (semantics). Communities can be a key to understanding the utility of a certain social network and why people join it (user intentions).

By analyzing microblogging communities in Twitter, we describe some of the user intentions that shed light on how people are participating in such platforms. Finally, it is typical that many blog posts are emotionally charged. The current models treat hyperlinks as endorsement. We describe how sentiment information around the link provides clues to it's polarity and can be used to identify influence and bias in social media. There are several applications that can benefit from these techniques: business intelligence, social recommendation, filtering tools and advertising; to name a few. Since social graphs are extremely huge and we are dealing with vast amounts of real-time data, we are exploring two approaches: one to develop efficient approximation approaches, and another that seeks to leverage the power of cell.