Event detection using Twitter: a spatio-temporal approach.

Area of Science:

• Social Media Analysis
• Data Mining
• Geospatial Statistics

Background:

• Twitter generates 400 million tweets daily, serving as a valuable data source for event detection.
• Current methods rely on keyword monitoring, requiring prior event knowledge and potentially missing optimal keywords.
• This limitation hinders comprehensive event analysis and monitoring.

Purpose of the Study:

• To introduce an alternative methodology for event detection using Twitter data.
• To evaluate the effectiveness of space-time scan statistics (STSS) for identifying event-related tweet clusters.
• To demonstrate STSS's capability in analyzing spatio-temporal patterns of information dissemination.

Main Methods:

• Utilized space-time scan statistics (STSS) to detect clusters in Twitter data across space and time.
• Applied the methodology to the 2013 London helicopter crash as a case study.
• Analyzed tweet data for emergent spatio-temporal clusters indicative of events, irrespective of content.

Main Results:

• Identified a spatio-temporally significant cluster corresponding to the London helicopter crash.
• The detected cluster, though transient, provided rich information including keywords and photographs.
• Successfully detected other events like football matches and transport delays using the STSS method.

Conclusions:

• Space-time scan statistics (STSS) is an effective technique for event detection in Twitter data.
• The method overcomes limitations of keyword-based approaches by focusing on spatio-temporal activity.
• STSS offers a robust framework for analyzing diverse real-world events from social media streams.