Go to ACL ARR 2025 February homepageExploring the Truth with Dialogue: Dual Large Language Model Interaction Cooperation and Multi-view Semantic Fusion Network for Fake News Detection
The widespread dissemination of fake news poses a significant threat to social trust and individual decision-making, necessitating advanced fake news detection technologies. Although integrating small language models (SLMs) with large language models (LLMs) has shown promise in detecting fake news, existing fake news detection methods with LLMs exploit large language models to generate extra knowledge of the social context for fake news detection. However, the LLMs themselves suffer from the hallucinations - generating plausible yet factually incorrect content. In addition, the SLMs of current methods focus on data consistency rather than data diversity when integrating multivariate information, resulting in incomplete information fusion. To address these challenges, we propose a novel fake news detection framework DLLM-MVSFN that combines a dual large language model interaction cooperation module and a multi-view semantic fusion network. DLLM-MVSFN leverages an interactive dialogue between two LLMs to generate comprehensive summaries of news events. Then a multi-view semantic fusion network is proposed to effectively integrate information from news content, LLMs summaries, and user comments for fake news detection. The experimental results show that our proposed DLLM-MVSFN outperforms existing baselines in multiple public datasets, achieving higher accuracy and F1 scores.