Do you believe it? Examining user engagement with fake news on social media platforms

Abstract

The proliferation of fake news on social media platforms makes it necessary to investigate how news content and user comments can influence user engagement. This study analyzes a robust dataset of 600 fake news posts on Facebook and 760,000 associated user reactions and comments. Employing topic modeling and regression reveals how content and social response characteristics interact to predict engagement. Analysis of textual, rhetorical, semantic, emotional, contextual, and source-based features provides a comprehensive methodology for modeling fake news dissemination. Results demonstrate multimedia inclusion, source credibility, ease of reading, political and technological topics, positive/anticipatory emotions, creator status, and comment deviation most strongly predict reactions, shares, and comments. The inclusion of 47 statistically significant interaction terms substantially improves regression fit and predictive accuracy. The random forest model achieves the highest cross-validation performance, demonstrating machine learning's capability to model fake news engagement's intricacies. These rigorous, data-driven findings provide important insights into engagement drivers and practical tools to mitigate fake news spread. The multidimensional feature set and predictive modeling approach provide a powerful methodology for decoding complex user-news dynamics. This study contributes to a better understanding of how fake news content and social contexts interact to engage users, empowering platforms, regulators, and researchers to counteract fake news.