Supporting Early and Scalable Discovery of Disinformation Websites

Online disinformation is a serious and growing sociotechnical problem that threatens the integrity of public discourse, democratic governance, and commerce. The internet has made it easier than ever to spread false information, and academic research is just beginning to comprehend the consequences. In response to this growing problem, online services have established processes to counter disinformation. These processes predominantly rely on costly and painstaking manual analysis, however, often responding to disinformation long after it has spread. We design, develop, and evaluate a new approach for proactively discovering disinformation websites. Our approach is inspired by the information security literature on identifying malware distribution, phishing, and scam websites using distinctive non-perceptual infrastructure characteristics. We show that automated identification with similar features can effectively support human judgments for early and scalable discovery of disinformation websites. Our system significantly exceeds the state of the art in detecting disinformation websites, and we present the first reported real-time evaluation of automation-supported disinformation discovery. We also demonstrate, as a proof of concept, how our approach could be easily operationalized in ordinary consumer web browsers.