Confined Twitter Conversations

The rise in popularity of social media websites has transformed internet users from passive consumers to active producers of digital content; enabling internet users to share their personal stories, recommendations, and opinions with a larger audience. In recent years, many businesses have availed of this trend to learn more about their customers' satisfaction as well as their product's uptake by exploiting user interaction data from popular social media websites (e.g., Twitter). However, not all broadcast information is valuable, and accurately extracting important discussions (i.e., information flow amongst users) and identifying important users has proven a particularly challenging task due to the magnitude of the available data and the large noise around user interactions. In this work we propose an approach for filtering and tracking complete conversations around targeted topics. Looking at individual conversations in this way has not been well studied in the literature. In particular we filter conversations from three different domains; Sport, Academia, and Politics. We model the extracted conversations as graphs and study why particular tweets become conversations (when many do not). We study the fundamental topological properties of these graphs and classify them using a landmark dK series. Finally we model the set of users in the conversation as another graph, identify the most common topologies and study methods for identifying influential users beyond the conversation starter.