Abstract

The advent of “social media” has enabled millions of people to participate in discussions within communities on a global scale. These conversations take place in a myriad of venues, on or off the web, each with its particular approach to implement what we now call “social media” – blogs, bulletin boards, mailing lists. However, while the software powering these communities varies a great deal, and continues to evolve, all of them share a common set of features. When a user initiates a discussion, the message is not addressed to a specific person, but broadcast to any interested reader; such a message can generate replies from other users, and these replies can then generate their own, forming a network of connections between messages. There is a need for a system that can make connections between related pieces of social text, to group information into coherent units. Making use of the structure of the social text helps to determine which elements of the text to consider for a given topic. To do this, a system needs to consider the different contexts in which it can be understood. A post, text transmitted by a single author at the same point in time, may have a different topic than the whole thread, which is comprised of all the posts in the discussion following an initial post. Different passages in a post could also have separate topics. Therefore, it is useful to annotate the text with information about its social structure explicitly for use in automatic search and text mining.