A Hybrid Attention Framework for Fake News Detection with Large Language Models
Xiaochuan Xu, Peiyang Yu, Zeqiu Xu, Jiani Wang
TL;DR
This work tackles the challenge of fake news detection by moving beyond text-only semantics to a hybrid framework that fuses statistical writing features (e.g., capitalization, punctuation, text length) with deep semantic representations from large language models. A two-layer, hybrid attention mechanism enables dynamic weighting and cross-feature interaction between statistical and semantic signals, capturing complex patterns that distinguish fake from real news. The approach yields a 1.5% improvement in F1 on the WELFake dataset (0.945 F1) over strong baselines and provides SHAP-based interpretability to aid content reviewers. These contributions offer a scalable, interpretable, and robust solution for automated fake news detection with potential applicability to real-time moderation and online information ecosystem integrity.
Abstract
With the rapid growth of online information, the spread of fake news has become a serious social challenge. In this study, we propose a novel detection framework based on Large Language Models (LLMs) to identify and classify fake news by integrating textual statistical features and deep semantic features. Our approach utilizes the contextual understanding capability of the large language model for text analysis and introduces a hybrid attention mechanism to focus on feature combinations that are particularly important for fake news identification. Extensive experiments on the WELFake news dataset show that our model significantly outperforms existing methods, with a 1.5\% improvement in F1 score. In addition, we assess the interpretability of the model through attention heat maps and SHAP values, providing actionable insights for content review strategies. Our framework provides a scalable and efficient solution to deal with the spread of fake news and helps build a more reliable online information ecosystem.