ZoFia: Zero-Shot Fake News Detection with Entity-Guided Retrieval and Multi-LLM Interaction

TL;DR

ZoFia tackles the challenge of dynamic fake news detection under LLM knowledge cutoffs by decoupling retrieval from judgment. It introduces entity-guided retrieval to fuse up-to-date external evidence with in-news content into a Multi-Source Information Matrix, followed by a dual-state multi-LLM interaction with role-specific agents and adversarial debate to produce interpretable judgments. Key innovations include Hierarchical Salience for robust entity importance, the Salience-Calibrated MMR (SC-MMR) for keyword selection, and a modular, debater-based reasoning chain that reduces hallucination and bias. Empirical results on GossipCop and PolitiFact show ZoFia outperforms most zero-shot and many few-shot baselines, with ablations highlighting the importance of retrieval and multi-agent collaboration for robustness and interpretability.

Abstract

The rapid spread of fake news threatens social stability and public trust, rendering its detection an imperative research priority. Although large language models (LLMs) excel at numerous natural language processing tasks with their remarkable contextual understanding and extensive prior knowledge, the time-bounded knowledge coverage and tendency for generating hallucination content reduce their reliability when handling fast-evolving news streams. Furthermore, models trained on existing static datasets also often lack the generalization needed for emerging news topics. To address these challenges, we propose ZoFia, a novel two-stage zero-shot fake news detection framework. First, we introduce Hierarchical Salience to quantify the importance of entities in the news content, and propose the SC-MMR algorithm to effectively select an informative and diverse set of keywords that serve as queries for retrieving up-to-date external evidence. Subsequently, a multi LLM interactive system, in which each agent assumes a distinct role, performs multi-view collaborative analysis and adversarial debate over the news text and its related information, and finally produces an interpretable and robust judgment. Comprehensive experiments on two public datasets demonstrate that ZoFia obviously outperforms existing zero-shot baselines and most of few-shot methods. Our codes will be open-sourced to facilitate related communities.

Figures (4)

• Figure 1: Overall architecture of our proposed ZoFia framework.
• Figure 2: The diagram of Multi-Source Information Matrix and the quadrants used by LLM agents.
• Figure 3: Performance comparison of ZoFia’s keyword extraction with other extraction methods.
• Figure 4: The effect of the number of keywords $k$ on performance (F1-Score).