Multi-order hyperbolic graph convolution and aggregated attention for social event detection.

Area of Science:

• Computer Science
• Data Science
• Artificial Intelligence

Background:

• Social event detection (SED) is crucial for public safety and marketing analytics.
• Conventional models struggle with the complex, hierarchical, and dynamic nature of social media data.
• Existing methods face challenges in capturing non-Euclidean relationships and higher-order event structures.

Purpose of the Study:

• To develop a novel framework for social event detection (SED) that addresses the limitations of Euclidean-based models.
• To effectively model heterogeneous, hierarchical, and dynamic social data structures.
• To improve the accuracy and robustness of event identification from social media streams.

Main Methods:

• Proposed the Multi-Order Hyperbolic Graph Convolution and Aggregated Attention (MOHGCAA) framework.
• Employed multi-order graph convolution within hyperbolic space.
• Integrated curvature-aware attention to capture local and global dependencies.

Main Results:

• MOHGCAA consistently outperformed state-of-the-art baselines in both supervised and unsupervised settings.
• Demonstrated robustness and scalability across multiple datasets.
• Showcased effectiveness in representing hierarchical and heterogeneous data structures.

Conclusions:

• The MOHGCAA framework provides a robust foundation for social event detection in non-Euclidean domains.
• Hyperbolic space and aggregated attention are effective for modeling complex social data.
• The study advances the capabilities of SED for real-world applications.