Beyond Content: Integrating Generated User Intent and Planned Behavior Theory for Robust Fake News Detection

ACL ARR 2025 February Submission1244 Authors

Abstract: The dissemination of true and fake news is often driven by distinct user motivations, yet existing detection methods predominantly focus on news content or propagation structures, often overlooking these underlying intents. This oversight can make such methods vulnerable to sophisticated adversarial strategies, such as crafted fake content or deceptive user engagement. While large language models (LLMs) provide rich, multi-dimensional behavioral insights, their standalone performance in detection often lags behind supervised models. To bridge this gap, we propose a novel computational framework that integrates the Theory of Planned Behavior (TPB) with LLM-generated user intent, enabling a deeper understanding of users' decision-making processes in news sharing. We employ a two-layer contrastive feature fusion mechanism to construct comprehensive behavioral representations, significantly enhancing fake news detection. Extensive experiments across four diverse datasets demonstrate that our method also exhibits remarkable robustness against adversarial attacks.

Paper Type: Long

Research Area: Computational Social Science and Cultural Analytics

Research Area Keywords: human behavior analysis, misinformation detection and analysis

Contribution Types: Model analysis & interpretability, Data analysis, Theory

Languages Studied: English, Chinese

Submission Number: 1244