Rumor detection model with weighted GraphSAGE focusing on node location.

Area of Science:

• Artificial Intelligence
• Social Media Analysis
• Natural Language Processing

Background:

• Social media facilitates rapid information spread, but also amplifies rumors.
• Traditional deep learning models struggle with the complex node relationships and information propagation dynamics in rumor detection.
• Existing methods often use fixed weights or mean aggregation, limiting accuracy and robustness.

Purpose of the Study:

• To develop an advanced rumor detection model that addresses limitations of traditional methods.
• To enhance the accuracy and robustness of rumor detection on microblogging platforms.
• To propose a location-aware weighted GraphSAGE rumor detection model (GSMA).

Main Methods:

• Introduced an attention mechanism for dynamic weighting of neighboring nodes during aggregation.
• Incorporated modulated position encoding to capture node position and order information.
• Integrated post text sentiment analysis to provide additional semantic features for rumor detection.

Main Results:

• The GSMA model achieved high accuracy rates of 97.43% on Ma-Weibo and 97.55% on Weibo23.
• Demonstrated an improvement of 1.11% and 0.77% over the benchmark GraphSAGE model, respectively.
• Showcased enhanced performance across all evaluation metrics compared to other state-of-the-art rumor detection models.

Conclusions:

• The proposed GSMA model significantly improves rumor detection accuracy and robustness.
• The integration of attention mechanisms, position encoding, and sentiment analysis is effective for social media rumor detection.
• GSMA offers a promising approach for timely and accurate identification of online rumors.