This article proposes a novel multi-agent system that emulates human annotator decision-making processes to detect and debunk disinformation, achieving superior results compared to individual Large Language Models like GPT-4 and GPT-3.5.
- The system incorporates consensus mechanisms, cognitive diversity, knowledge diversity, and hierarchical structures inspired by human behavior.
- It leverages open-source models including LLaMA, Kimi, Qwen, Deepseek, and LLaMA-Nemotron to ensure transparency.
- Evaluation covers datasets in English (high-resource), Polish (medium-resource), Slovak (low-resource), and Bulgarian (low-resource).
- Experiments address direct disinformation detection, identification of texts worthy of verification, and detection of verifiable factual claims.
The authors consider this approach important because the scale of disinformation renders manual fact-checking inadequate, necessitating automated methods that outperform single LLMs while maintaining transparency through open-source components.