Authors: Logan Walls, Jim Maddock, Kate Starbird and Emma Spiro
Social media are important for rapidly obtaining and disseminating information, particularly when information is time-sensitive and critical for decision making. However, as individuals try to make sense of information in non-routine situations, rumors and misinformation often proliferate, eroding users’ confidence in these platforms (Turner, 1994; Hill, 2012; Hiltz et al., 2014; Hughes & Palen, 2012). As such, empirical studies of rumoring on social media—and informal communication dynamics more generally—have become a highly relevant area of study. One challenge facing such work is determining a method of identifying relevant content (e.g. tweets, posts) within the larger social media stream. Our work directly addresses this problem, using unsupervised learning methods to augment qualitative approaches to define datasets of interest within a large corpus of text-based data.
