Propagation Structure Fusion for Rumor Detection Based on Node-Level Contrastive Learning.

Area of Science:

• Computer Science
• Social Computing
• Artificial Intelligence

Background:

• Online rumor propagation poses significant societal and economic risks.
• Existing graph contrastive learning models for rumor detection face limitations in distinguishing semantically similar propagation structures.
• This necessitates advanced methods to improve the accuracy and robustness of rumor detection systems.

Purpose of the Study:

• To propose a novel propagation fusion model based on node-level contrastive learning (PFNC) for enhanced online rumor detection.
• To address the issue of negative samples with similar structures degrading model performance in existing contrastive learning approaches.
• To improve the discriminative power and overall effectiveness of rumor detection models.

Main Methods:

• PFNC generates three augmented propagation structures via node text masking and edge perturbation.
• Node-level contrastive learning is applied between augmented structures to preserve similarities.
• A CNN-based model treats augmented structures as color channels for information fusion.

Main Results:

• The proposed PFNC model demonstrates significant performance improvements over state-of-the-art methods.
• Experimental results on real-world datasets validate the effectiveness of the PFNC approach.
• The method successfully prevents the misclassification of semantically similar propagation structures.

Conclusions:

• PFNC offers a robust and effective solution for online rumor detection by refining graph contrastive learning.
• The fusion of augmented propagation structures captures crucial information for accurate classification.
• This research contributes to mitigating the negative impacts of online rumors.