Social Media Analytics for Disaster Management

In times of crisis involving disasters or other extreme events, victims of these events use social media to
share information about their situation. The user-generated content contains vast amounts of valuable
information, albeit mostly hidden, regarding the victims’ needs, the urgency of supplies, and their
situation following the disaster. The insights are quite valuable for humanitarian operations to develop
a real time understanding of the situation even before they arrive at the ground. Moreover, time is
an important dimension in humanitarian relief activities, since providing assistance and supplies to
victims in a timely manner helps save lives and mitigate the effects of the disaster. With this in mind, in
addition to exploring methodological aspects related to identifying and extracting real-time information
and analyzing disaster-related social media data through a theoretical lens, this dissertation contributes
to the understanding of the importance of extracting real-time information in the context of a disaster.
Our literature review revealed that there has been very little attention devoted to the use of social
media in disasters in information systems (IS) research. For the empirical part of this dissertation, we
focused on two different natural catastrophes: For Study 1, we applied a manual content analysis
method to analyze the social media data related to Hurricane Sandy through the theoretical lens of
situation awareness. For Study 2, we applied a supervised machine learning approach to analyze
the media data related to the Chennai floods through the theoretical lens of social presence. For
Study 3, we applied the unsupervised topic modeling method to analyze social media data related
to the Chennai floods for the purpose of understanding the emerging phenomena during disasters.
Furthermore, inductively derived reflections and findings based on the empirical studies (Study 1 and 2)
further helped clarify the challenges and opportunities associated the methods point of view and social
media data point of view. Furthermore, we developed a time-indicating dictionary—a time wordlist
(T-wordlist) to automatically process and extract time-relevant information of needs and urgencies
during disasters from social media content. This research extends a social media analytics method—the
dictionary-based approach—by developing the T-wordlist to extract the time-relevant information from
the social media content.
Overall this dissertation contributes to current research in two ways. First, research findings confirm
the theoretical explanations of situation awareness and social presence and shed light on emerging
phenomena. Most importantly, this dissertation extends the methodological aspects by analyzing the
theoretical concept of social presence, since thus far this concept has only been analyzed through
survey strategy or by conducting interviews. The research at hand extends this concept further by
analyzing Twitter messages and examining why people choose to help one another in times of crisis
despite the lack of a personal relationship. Second, the contributions of this dissertation have important
implications for practitioners. Since disaster management agencies must handle massive volumes of
data during disasters, the T-wordlist could be embedded into their systems, resulting in an automatic
extraction of time-relevant information concerning needs and urgencies.