Nowhere to Hide: Online Rumor Detection Based on Retweeting Graph Neural Networks.

Area of Science:

• Computer Science
• Artificial Intelligence
• Social Media Analysis

Background:

• Online rumor detection is vital for a healthy digital environment.
• Traditional content-based methods are insufficient due to content manipulation.
• Information propagation patterns offer richer insights but are complex to model.

Purpose of the Study:

• To develop a novel rumor detection framework leveraging propagation patterns.
• To address the topological complexity of retweeting trees.
• To integrate content, user, and structural information for enhanced detection.

Main Methods:

• Proposed a structure-aware retweeting graph neural network framework.
• Developed a conversion method to transform complex retweeting trees into binary trees.
• Serialized trees into meta-tree paths for deep neural network integration.
• Integrated content, user features, and structural embeddings via self-attention and mutual attention mechanisms.

Main Results:

• The proposed model demonstrated superior performance on two real-world datasets.
• Effectively captured complex propagation patterns through graph neural networks.
• Achieved more reliable rumor detection by fusing diverse data sources.

Conclusions:

• Structure-aware graph neural networks offer a powerful approach for online rumor detection.
• Integrating propagation patterns significantly enhances detection accuracy over content-based methods.
• The novel framework provides comprehensive representations for robust rumor detection.