Theory based features improve fake news detection and explanation
Fake News Theories: Harnessing Disciplinary Insights for Computational Modeling, Detection, and Explanation
Machine LearningComputers and Society
Summary
Detecting fake news is hard because many detection tools don't connect well to how people actually believe or spread false information. The authors review ideas from psychology, social science, and economics to find measurable clues about fake news persuasion. They use these clues to build models that not only detect fake news better but also help explain why something is likely fake. Their tests show that combining these theory-based clues gives better and more understandable results.
What this means in practice
- •For social media platform engineers: Incorporate theory-based features to improve fake news detection while providing explanation signals to moderation teams.
- •For news verification service providers: Use interpretable, theory-linked detection features to produce explanations alongside automated fake news flags for fact-checkers.
Authors
Zhaoyang Cao, Miriam Metzger, Reza Zafarani
Abstract
Disinformation research has produced increasingly accurate automated fake-news detectors, but many systems remain difficult to interpret and are weakly connected to established theories of persuasion, credibility, and human judgment. In this paper, we develop a theory-informed computational framework that translates cross-disciplinary theories of fake news into measurable features for automated detection and explanation through statistical techniques and large language models. To that end, we conduct a structured cross-disciplinary review of theories from social sciences, psychology, economics, among other disciplines that reveal how fake news persuades and spreads, thereby establishing a broad theoretical foundation for computational modeling. Experiments on benchmark datasets show that theory-derived features are predictive and provide interpretable, theory-referenced diagnostic signals. Multi-feature models generally outperform individual features, although gains among the strongest small feature combinations are modest. Our work highlights the value of interdisciplinary perspectives in building robust and interpretable fake news detection systems, advancing the foundation for human-centered approaches in combating disinformation.