Automatic Identification of Crisis-Related Sub-Events using Clustering

Daniela Pohl, University of Klagenfurt

Abdelhamid Bouchachia, Bournemouth University

Hermann Hellwagner, University of Klagenfurt



Social media are becoming an important instrument for supporting crisis management, due to their broad acceptance and the intensive usage of mobile devices for accessing them. Social platforms facilitate collaboration among the public during a crisis and also support after-the-fact analysis. Thus, social media are useful for the processes of understanding, learning, and decision making. In particular, having information from social networks in a suitable, ideally summarized, form can speed up such processes. The present study relies on Flickr and YouTube as social media and aims at automatically identifying individual sub-events within a crisis situation. The study applies a two-phase clustering approach to detect those sub-events. The first phase uses geo-referenced data to locate a sub-event, while the second phase uses the natural language descriptions of pictures and videos to further identify the ”what-about” of those sub-events. The results show high potential of this social media-based clustering approach for detecting crisis-related sub-events.



Proceedings of the International Conference on Machine Learning and Applications (ICMLA 2012),

Boca Raton, Florida, USA, 12-15 December 2012.



Upcoming Events

Read about BRIDGE and other events in the field of disaster management We will keep you informed about the workshops, conferences, exhibitions, and exercises that might be of interest to you.
Learn more

Recent Deliverables

D06.1 - Design Pattern Library - is ready for browsing! Please visit the BRIDGE Design Pattern Library and submit new patterns at
Learn more

Recent Publications

International Cooperation in Crisis Management – A European Perspective Large scale crises can influence many countries simultaneously as they may occur in or involve multinational regions.
Learn more