Related Experiment Videos
Beyond semantics: Exploiting propagation structures with dual-adapter LLMs for fake news detection
Aojie Si1, Shizhan Chen1, Xiaobao Wang1
1College of Intelligence and Computing, Tianjin University, No. 92 Weijin Road, Tianjin, 300072, China.
Summary
This study introduces Propagation Structure-Augmented LLMs (PSALLM), a new method for detecting fake news by analyzing how information spreads. PSALLM effectively identifies fake news across various domains, outperforming existing methods.
Area of Science:
- Artificial Intelligence
- Natural Language Processing
- Computational Social Science
Background:
- The proliferation of fake news across diverse platforms threatens public opinion and societal stability.
- Traditional fake news detection methods, reliant on content analysis, are increasingly ineffective due to large language models (LLMs) facilitating fake news generation.
- There is a critical need for advanced, generalized detection techniques capable of addressing complex, real-world scenarios.
Purpose of the Study:
- To propose a novel fake news detection model, Propagation Structure-Augmented LLMs (PSALLM), that integrates news propagation structures with LLMs.
- To enhance fake news detection by leveraging the understanding of how information propagates across networks.
Main Methods:
- PSALLM utilizes pre-trained LLMs augmented with two specialized adapters: an adapter with non-weight-decay (ANWD) for invariant structural features and an adapter with weight-decay (AWD) for domain-specific structural shifts.
- An adaptive distillation strategy, based on domain similarity, is employed to dynamically balance the contributions of both adapters.
- The model is designed to perceive and analyze news propagation structures for improved detection accuracy.
Main Results:
- PSALLM demonstrated superior performance compared to state-of-the-art baselines across four real-world cross-domain datasets.
- The model effectively handles diverse news domains and various manipulation strategies, showcasing its robustness.
- Experimental results confirm the effectiveness of leveraging propagation structures within LLMs for fake news detection.
Conclusions:
- PSALLM represents a significant advancement in fake news detection, offering a more robust and generalized approach.
- The integration of propagation structures with LLMs provides a powerful framework for identifying sophisticated fake news.
- The proposed method shows promise for real-world applications in combating misinformation across different domains.
