Understanding Digital Misinformation Across Platforms and Modalities

Abstract

This dissertation investigates how misinformation spreads across digital platforms and what factors amplify its reach. It examines the role of opaque algorithmic systems—such as those used by YouTube and Amazon—in shaping user exposure to false or misleading content. It also explores how narrative techniques used in social media posts influence public engagement with misinformation, and how manipulated videos, especially those deceptively edited, can be detected using advanced machine learning models. The research is based on four complementary projects that together provide a multi-layered analysis: two algorithmic audits of recommendation systems, a large-scale study of narrative styles in health-related tweets, and a multimodal framework for detecting deceptive video content. Collectively, these studies contribute new datasets, methods, and insights toward understanding and combating misinformation in modern information ecosystems.