LADKG: LLM-Augmented Dynamic Knowledge Graph for Fake News Detection

ACL ARR 2026 January Submission3796 Authors

Keywords: Information Extraction, Information Retrieval and Text Mining, NLP Applications

Abstract: Fake news spreads rapidly on social media and causes serious societal harm. Existing methods rely on propagation structures, temporal signals, or external knowledge, but often model them separately and fail to capture dynamic diffusion, evolving comment-based knowledge, and high-order semantic relationships. We propose llm-augmented dynamic knowledge graph (LADKG), a unified framework for fake news detection. LADKG constructs a dynamic post–entity–concept knowledge graph from posts and user comments using large language models and updates it over time to capture semantic evolution. A multi-hop graph attention mechanism aggregates high-order neighborhood information for deep semantic reasoning. LADKG further introduces a post-enhancement unit to model fine-grained interactions between textual and knowledge representations. Experiments on two datasets show that LADKG consistently outperforms strong baselines, with notable gains in early-stage fake news detection.

Paper Type: Long

Research Area: NLP Applications

Research Area Keywords: Information Extraction, Information Retrieval and Text Mining, NLP Applications

Contribution Types: Model analysis & interpretability, NLP engineering experiment, Reproduction study, Publicly available software and/or pre-trained models, Data resources, Data analysis, Position papers, Surveys, Theory

Languages Studied: English

Submission Number: 3796