Event detection via graph convolutional networks for multi-source multimodal integration

Area of Science:

• Computer Science
• Data Science
• Artificial Intelligence

Background:

• Social media platforms generate vast amounts of real-time data crucial for event detection.
• Existing event detection methods often rely on single data sources or modalities, limiting their effectiveness.
• The complex, multimodal nature of event information requires integrated approaches.

Purpose of the Study:

• To develop a novel framework for robust real-time event detection.
• To integrate multi-source and multimodal data for enhanced event understanding.
• To improve the accuracy and reliability of event detection and classification.

Main Methods:

• Utilized graph convolutional networks to capture textual and visual information, including syntactic and dependency details.
• Employed cross-modal attention mechanisms to effectively combine features from text and images.
• Developed a model to identify event trigger words and classify events using combined textual and visual cues.

Main Results:

• The proposed framework demonstrated improved event detection accuracy and reliability.
• Integrating text, images, and news data led to significant increases in event detection and classification performance.
• The multi-source, multimodal approach offers a more dynamic and comprehensive solution for event detection.

Conclusions:

• Multi-source, multimodal data integration is essential for advanced real-time event detection.
• Graph convolutional networks and cross-modal attention effectively process complex event information.
• The framework provides a robust and accurate solution for identifying and classifying real-world events.