Article Preview
Top1. Introduction
During the occurrence of disasters, may it be natural or humanmade, people use social media excessively compared to normal hours. Floods, earthquake, tsunamis, stampede, fire, terrorist attack etc. are a couple of events that may lead to significant loss of lives and property. At such hours, social media platforms, such as Twitter and Facebook, become a primary source of information (Imran et al., 2015). These platforms give people the means of interaction with their family and friends. During such events, people usually share information about themselves and also query about their family and friends. This act of continuous and real-time posting of status and messages on these platforms results into accumulation of social interaction data, which can be further processed to extract useful information that helps in disaster response. The people contributing the messages on such platforms during disastrous hours are either those who have experienced the event on the ground or those who have heard something about the ongoing event. Thus, the information retrieved from those messages can be beneficial for the first responders as well as the decision makers to come up with effective plans.
Previous studies (Abel et al., 2012; Ashktorab et al., 2014; Caragea et al., 2011; Imran et al., 2014; MacEachren et al., 2011; Purohit & Sheth, 2013; Sheth et al., 2010; Vieweg et al., 2010; Yin et al., 2015) have shown that information extracted out of the tweets which are posted publicly on the Twitter platform contribute to better understanding of an event, as it unfolds, and also provides a direction to develop actionable plans (Antoniou & Ciaramicoli, 2013). Twitter possesses real-time and extremely informal communication characteristics. Unlike other social platforms, Twitter has character limitations for each tweet, which is why people tend to use Twitter as an informal medium to express their status. Processing such small messages (tweets) over spatial, temporal and thematic dimensions requires researchers to overcome a couple of challenges. Twitter, however, contribute to emergency situational awareness and has provided data for the researchers. Besides, Twitter also provides developers with multiple APIs to access the data available on its platform—Streaming API and Search API are the ones used extensively for streaming tweets locally (Lamsal, 2021).
During a disastrous event, it is reasonable for the number of tweets to rise exponentially (Yin et al., 2015). Millions of tweets relating to disasters such as Nepal Earthquake 2015, India Floods 2014, Ebola Virus Outbreak 2014 have been collected in the past (Imran et al., 2016), including the ongoing COVID-19 pandemic (Lamsal, 2020). It is practically not possible for human beings to manually process and analyze each tweet. Artificial intelligence (AI) techniques have been widely used to address similar tasks (Lamsal & Vijay Kumar, 2020b, 2020c). In this paper, a semi-automated AI-based disaster response system is proposed, whose core part is a machine learning classification model that classifies an incoming tweet into categories related to community needs, lives lost, damages caused etc. The categorized tweets can be sent to the respective government bodies, and can be further processed to understand the spatial, temporal and semantical dimensions related to the event. Further, a tentative estimation of the critically affected area can be sketched on the map using the classified tweets.
The paper is organized as follows: A disaster response system is proposed in Section 2 and the methodology is discussed in detail in Section 3. Experimentation and results are shown in Section 4 followed by conclusion in Section 5.