Social media as intelligence in disaster response: eyewitness classification using community detection
<p>Disasters cause widespread devastation to both physical infrastructure and the lives of individuals residing in large geographic areas. The disruption caused by disaster events is further compounded by high levels of uncertainty and information scarcity, presenting significant challenges to...
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2022
|
| Subjects: |
| Summary: | <p>Disasters cause widespread devastation to both physical infrastructure and the lives of individuals residing in large geographic areas. The disruption caused by disaster events is further compounded by high levels of uncertainty and information scarcity, presenting significant challenges to disaster response organisations and impeding the effectiveness of coordinated response efforts.</p>
<br>
<p>The increasing use of digital technologies, such as social media, presents valuable sources of information that are available in real-time from geographically-distributed networks of ‘humans as sensors’. The data generated by these technologies can supplement traditional sources of intelligence to build models of situational awareness and inform decision-making, resulting in more effective disaster response operations.</p>
<br>
<p>This thesis proposes a method of curating social media data to enhance its usefulness as a source of intelligence for disaster response organisations during crisis events. The research was conducted in four phases:</p>
<p>(i) An ethnographic study developed a conceptual framework of the values and challenges of social media intelligence as perceived by disaster response practitioners. High data volume and low rates of relevance were established as key factors impeding integration with existing intelligence sources.</p>
<p>(ii) Empirical studies of Twitter discourse were conducted during eight disaster events to identify patterns of online behaviour and establish the informative potential of social media data as a rich source of eyewitness reports.</p>
<p>(iii) Geoproximate preferential attachment (homophily) was identified in the structure of Twitter relationship networks. An eyewitness classification model integrated relationship features for data curation. The model was evaluated on temporally-partitioned subgraphs and shown to be effective in real-time environments.</p>
<p>(iv) The classification model was validated in simulated disaster response scenarios conducted with emergency service practitioners. Feedback from participants confirmed the effectiveness of the approach to improving the practical value of social media data as a source of intelligence during disaster response operations.</p> |
|---|