DGFND-ES: Evidence-Enhanced Dual-Graph Fake News Detection under Sociological Constraints

ACL ARR 2026 January Submission3745 Authors

Keywords: Multimodal Fake News Detection, Neuro-Symbolic Framework, Dual-Graph, Sociological Constraints, Evidence Retrieval

Abstract: The coordinated dissemination of multimodal content on social media has become the norm, rendering fake news increasingly covert and complex. Existing methods generally lack event-level information modeling, which limits their ability to effectively handle breaking events, and they also fail to account for group fairness and disparities in information harm under a globalized context. To address these challenges, we propose DGFND-ES, a dual-graph collaborative fake news detection framework that integrates evidence enhancement with sociological constraints. This framework adopts a neural-symbolic architecture consisting of a “main graph–consistent subgraph” structure, and incorporates group fairness constraints and a harm-aware loss during training to endow the model with social responsibility. In addition, we construct a high-quality SSS dataset for systematic evaluation of model performance. Experimental results demonstrate that DGFND-ES consistently outperforms existing methods on the Weibo-21, Fakeddit, and SSS datasets.

Paper Type: Long

Research Area: Computational Social Science, Cultural Analytics, and NLP for Social Good

Research Area Keywords: misinformation detection and analysis, language/cultural bias analysis, quantitative analyses of news and/or social media

Contribution Types: NLP engineering experiment, Data resources, Data analysis

Languages Studied: English, Chinese, Spanish, French, Russian

Submission Number: 3745