Determining levels of urgency and anxiety during a natural disaster: Noise, affect, and news in social media

Since 2010, and perhaps before that as well, news and views of and about citizens caught up in a natural disaster, like floods and hurricanes, are increasingly available through digital media channels. In social media via Twitter for instance- and in formal media, especially in the blogs accompanying news compiled by various public and private sector agencies, one can get information about events as they unfold. Monitoring this stream of digital information provides valuable information for rescue agencies. However, caution has to be exercised in that this stream of information can be ostensibly stored for future analysis, by say resilience planners, without due care for the privacy of named entities, including individuals, places and institutions. In this paper we present a scalable bag-of-words method for analysing social media and crowd-sourced documents to visualise the evolving signature of a disaster event comprising disaster and affect terms. We illustrate our method by using a hurricane and an earthquake case study and two systems developed at Trinity College Dublin an ontology-based, scale-oriented system called Rocksteady and a terminology and ontology extraction system called CiCui. Ethical questions raised by automatic collection and analysis of social media data, especially the collation and storage of named entities is discussed.