Rumor detection driven by graph attention capsule network on dynamic propagation structures.

Area of Science:

• Computer Science
• Artificial Intelligence
• Social Media Analysis

Background:

• Rumor detection on social media is crucial for preventing misinformation spread.
• Existing graph neural network methods struggle to capture deep-level rumor properties due to limited node features.
• Dynamic and interactive features in rumor propagation are often overlooked.

Purpose of the Study:

• To propose a novel model, the Graph Attention Capsule Network on Dynamic Propagation Structures (GACN), for enhanced rumor detection.
• To address the limitations of current methods in capturing deep-level and dynamic rumor characteristics.
• To improve the accuracy and effectiveness of identifying authentic versus rumored posts.

Main Methods:

• Developed GACN, integrating a capsule network with a graph attention network to encode static graphs into substructure classification capsules.
• Implemented a dynamic network framework to segment rumor structures into chronological static graphs, capturing evolving interactive features.
• Utilized a capsule attention mechanism to aggregate information from substructures, prioritizing informative ones.

Main Results:

• The proposed GACN model demonstrated superior performance compared to existing baseline methods.
• The model effectively mined deep-level properties and dynamic interactive features of rumors.
• Validation on two real-world datasets confirmed the effectiveness of the GACN approach.

Conclusions:

• GACN offers a significant advancement in rumor detection by leveraging capsule networks and dynamic graph analysis.
• The model's ability to capture both static substructure properties and dynamic propagation patterns is key to its success.
• GACN provides a more robust framework for combating the spread of online rumors.