Social Media Visual Analytics

Team Members: Mengdie Hu, John Stasko

     Visualization of textual content from social media and online communities - Hu PhD thesis 2018
     Visualizing Social Media Content with SentenTree - InfoVis 2016/TVCG 2017
     OpinionBlocks: A Crowd-Powered, Self-improving Interactive Visual Analytic System for Understanding Opinion Text - INTERACT 2013
     Breaking News on Twitter - CHI 2012

    VIDEO: Visualizing Social Media Content with SentenTree, Infovis '16 (39 MB mp4)
    DEMO & SOURCE CODE: Demos examples showing SentenTree running on different Tweet collections such as some from the opening game of the 2014 soccer World Cup (Code Repository)

Social media has become a significant part of modern society. It has been used to distribute press releases, drive marketing campaigns, and even start revolutions. Social media introduces a novel way of communication and produces "big data" that often baffle analysts who try to understand and utilize social media. This research project focuses on studying new analysis approaches, especially visualization techniques, that help improve our understanding of social media.

One of our early research efforts examines the leaking of Osama Bin Laden's death on Twitter in 2011. We reveal interesting facts about how breaking news works on social media. This project has been covered by Georgia Tech News Room, ReadWriteWeb, MSNBC, LA Times, Washington Times, American Public Media, Poynter Institute, and a number of other media outlets.

Our current research is developing visualization techniques that can summarize large collections of social media posts. We have developed the SentenTree technique that helps a person understand the main ideas, themes, and thoughts from collections of thousands of tweets. A SentenTree visualization is a little like a cross between a word cloud and a WordTree visualization. It shows the key words used a lot in the collection, but seeks to weave them together into more coherent sentence segments.

This research is supported in part by an IBM PhD Fellowship, by the DARPA XDATA project, and and by the National Science Foundation via award IIS-1320537.