Rumor detection based on propagation graph neural network with attention mechanism

Area of Science:

• Social Media Analysis
• Computational Social Science
• Machine Learning

Background:

• Social media rumors pose significant threats to social security.
• Existing rumor detection methods often overlook temporal dynamics and propagation patterns.
• Effective rumor detection requires advanced techniques considering information spread.

Purpose of the Study:

• To develop a novel representation learning framework for rumor detection.
• To address the limitations of existing methods by incorporating propagation structure.
• To enhance the accuracy and timeliness of identifying online rumors.

Main Methods:

• Constructing a propagation graph based on user reply structures on Twitter.
• Proposing a gated graph neural network (PGNN) for node representation learning.
• Developing GLO-PGNN and ENS-PGNN models utilizing global embedding and ensemble learning with attention mechanisms.

Main Results:

• PGNN effectively generates powerful node representations by exchanging information between neighbors.
• GLO-PGNN and ENS-PGNN models demonstrated superior performance compared to state-of-the-art methods.
• The proposed models achieved significant improvements in both rumor detection and early detection tasks on a real-world Twitter dataset.

Conclusions:

• The proposed PGNN-based framework offers a powerful approach for rumor detection.
• Incorporating propagation patterns and temporal dynamics is crucial for effective rumor detection.
• The GLO-PGNN and ENS-PGNN models represent a significant advancement in combating social media rumors.