Reproduce “Agentic Multi-Persona Framework for Evidence-Aware Fake News Detection” on OpenResearch

About this paper

Roopa Bukke, Soumya Pandey, Suraj Kumar, Soumi Chattopadhyay, Chandranath Adak

Researchers from the Indian Institutes of Technology Indore and Patna developed AMPEND-LS, an agentic multi-persona framework for evidence-aware fake news detection that integrates multimodal evidence retrieval, structured LLM reasoning, and SLM classification. The framework consistently outperformed state-of-the-art baselines across multiple datasets, achieving up to 6.17% higher accuracy and 9.07% higher F1 score on the GossipCop dataset, while also providing verifiable and informative justifications.

Key insights

• Multi-persona agentic reasoning, where diverse expert roles iteratively analyze evidence, significantly enhances the reliability and accuracy of fake news detection compared to single-persona or simpler approaches.
• The integration of a persuasion analysis module is crucial for resolving rhetorically manipulative and initially 'uncertain' cases, demonstrating substantial performance improvements when complementing evidence-based reasoning.
• Distilling the deep, evidence-grounded reasoning of Large Language Models into a lightweight Small Language Model classifier offers a pragmatic balance between high performance, computational efficiency, and scalability for real-world deployment.

Summary by alphaXiv.