Discovery of False Information on Digital Communication Channels

From rumors in ancient marketplaces to propaganda in print media, misinformation has always been part of human soci- eties. What has changed in the modern era is the scale, speed, and reach with which false information spreads. The major contribution in this spread is the accessibility and immediacy of modern communication channels. Unauthorized content is uploaded and widely broadcast, often being absorbed and believed by unsuspecting users. This issue jeopardizes public trust, influences their perceptions, and in severe cases, endan- gers their health. Robust automatic systems are essential to address this prob- lem. Therefore, a wide range of advanced methods is being explored by researchers for fake news detection on online so- cial networks. However, despite the considerable progress achieved, several limitations still exist. These challenges pri- marily relate to the inadequacy of updated datasets, the lack of integration of contextual information, the insufficient ap- proaches for detecting and controlling fake news on social networks, and the absence of evaluation of system resilience. This thesis aims to bridge these gaps by providing effective and practical solutions. First, a comprehensive and up-to- date dataset for fake news detection is presented, enriched with detailed information to support further research. Sec- ond, an efficient system for fake news detection is proposed, utilizing the strength of ensemble learning while integrating domain knowledge of news. Third, an innovative approach is proposed to limit the dissemination of fake news on online social networks. Lastly, it analyzes the strength of existing fake news detection models through a novel adversarial at- tack architecture.