Identifying Event-Specific Sources from Social Media

Abstract

Social media has become an indispensable resource for coordinating various real-life events by providing a platform to instantly tap into a huge audience. The participatory nature of social media creates an environment highly conducive for people to share information, voice their opinion, and engage in discussions. It is not uncommon to find novel and specific information with intimate details for an event on social media platforms in contrast to the mainstream media. This makes social media a valuable source for event analysis studies. It is, therefore, of utmost importance to identify quality sources from these social media sites for understanding and exploring an event. However, due to the power law distribution of the Internet, social media sources get buried in the Long Tail. The overwhelming number of social media sources makes it even more challenging to identify the valuable sources. We propose an evolutionary mutual reinforcement model for identifying and ranking highly ‘specific’ social media sources and ‘close’ entities related to an event. Due to the absence of ground truth, we provide a novel evaluation strategy for validating the model. By considering the top ranked sources according to our model, we observe a substantial information gain (ranging between 25 and 130 %) as compared to the baselines (viz., Google search and Icerocket blog search). Moreover, highly informative sources are ranked much higher according to our model as compared to the widely-used baselines, putting spotlight on the social media sources that could be easily overlooked otherwise. Our model further affords an apparatus to analyze events at micro and macro scales. Data for the research is collected from various blogging platforms such as, Blogger (hosted at blogspot), LiveJournal, WordPress, Typepad, etc. and will be made publicly available for researchers.