Deep-Eware: spatio-temporal social event detection using a hybrid learning model.

Area of Science:

• Social media event detection
• Big data analytics
• Geospatial information systems

Background:

• Social media generates vast amounts of data, offering opportunities for real-time event detection.
• Existing methods often struggle with the scale and complexity of social data streams.
• Accurate extraction of spatio-temporal event information remains a challenge.

Purpose of the Study:

• To develop a hybrid approach for extracting and clustering social events from big data streams.
• To present Deep-Eware, an efficient platform for real-time spatio-temporal event discovery and dissemination.
• To enable advanced smart city applications through enhanced event detection.

Main Methods:

• Utilized a hybrid learning model combining supervised deep learning (CNN, bidirectional LSTM) for feature extraction and topic classification.
• Employed unsupervised spatial clustering (hierarchical density-based) for event location inference.
• Integrated KeyBERT for semantic keyword generation and developed incremental machine learning algorithms for event discovery.

Main Results:

• Demonstrated the effectiveness and efficiency of the Deep-Eware platform using Twitter datasets.
• The hybrid approach significantly improves real-time spatio-temporal event detection and tracking.
• Achieved accurate event classification and location inference.

Conclusions:

• The Deep-Eware platform offers a scalable and efficient solution for social event detection.
• This hybrid approach provides a major advantage for real-time analysis of social media data.
• Enables novel smart city applications including event-enriched planning and emergency management.