A Hierarchical Graph-Language Model for Detection of Malicious Rumors in Social Media

View at publisher Summarise with VIRAT
Authors: Jianbin Wang, Yujie Lin, Qitao Tan, C. L. Philip Chen, Yue Su
Cited by: 0
References: 14

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

With the widespread adoption of social media, online rumors spread at an alarming rate, posing a serious threat to public perception and social stability. To address this, this paper proposes a Hierarchical Graph Language Model (HGLM) based on Graph Attention Networks (GAT) and Large Language Models (LLM) for early detection of online rumors. HGLM captures multi-layered semantic features of posts and their reply chains through hierarchical text modeling. It dynamically adjusts the influence between nodes in the propagation graph via an adaptive edge weight mechanism and employs multi-model expert fusion to achieve collaborative discrimination of text semantics and propagation structure. Experimental results demonstrate that on the Twitter15, Twitter16, and PHEME datasets, HGLM significantly outperforms traditional machine learning models and deep graph neural network models across metrics including Accuracy, F1 true, and F1 false. It exhibits outstanding stability, versatility, and cross-dataset transferability. This approach provides an efficient, scalable, and robust solution for early rumor identification within complex social network environments. Experimental results on Twitter15, Twitter16, and PHEME datasets demonstrate the model’s superiority. HGLM achieves accuracy rates of 89.2%, 89.1%, and 86.8% respectively, outperforming state-of-the-art baselines by approximately 6–9%. These results validate HGLM as a robust solution for early rumor identification, significantly advancing the capability to maintain social stability in complex network environments.

Fields of Study & Subjects

Misinformation and Its Impacts Complex Network Analysis Techniques Spam and Phishing Detection

Publication details

Journal / source
International Conference on Artificial Intelligence and Social Network Systems (AISNS)
Pages
88–91
Publication date
9 January 2026
Language
EN