A Rumor Detection Model Incorporating Propagation Path Contextual Semantics and User Information

Area of Science:

• Computer Science
• Artificial Intelligence
• Social Media Analysis

Background:

• Rumor detection is crucial for mitigating misinformation spread on social media.
• Existing methods often overlook critical structural and user-specific features in propagation patterns.
• This leads to suboptimal performance in identifying and flagging false information.

Purpose of the Study:

• To develop advanced models for more accurate and efficient rumor detection.
• To address the limitations of current methods by incorporating structural and user information.
• To improve early detection capabilities for online rumors.

Main Methods:

• Proposed a Dual-Attention Network on propagation Tree structures (DAN-Tree) utilizing a node-and-path dual-attention mechanism.
• Implemented path oversampling and structural embedding to enhance the learning of deep propagation structures.
• Introduced DAN-Tree++ by integrating user profiles into the propagation trees for enhanced performance.

Main Results:

• DAN-Tree demonstrated superior performance compared to state-of-the-art models on four rumor datasets based on propagation structures.
• DAN-Tree++ significantly outperformed other models when incorporating both user profiles and propagation structures on two datasets.
• Both DAN-Tree and DAN-Tree++ achieved top performance in early rumor detection tasks.

Conclusions:

• The proposed DAN-Tree and DAN-Tree++ models offer significant advancements in rumor detection.
• Integrating structural and user-specific features is vital for improving model accuracy and early detection.
• These models provide a robust framework for combating the spread of online misinformation.