Content area
Full text
1. Introduction
The automatic identification of information in disaster management (DM) is one of the most compelling global problems, as countries worldwide have witnessed a significant increase in the intensity of disasters. With the spiraling coronavirus disease, i.e. COVID-19, it has become indispensable to extract and disseminate accurate and timely information. The World Health Organisation declared the coronavirus disease outbreak as a pandemic on March 11, 2020.
The situation worsened day by day, and therefore, speedy and on-time information retrieval is imperative. Social distancing, lockdowns, travel bans, self-quarantines and business closures have forced people to glue to social media (SM) more than ever before (Zhang et al., 2020). SM’s real-time data production capability makes data enormous and diverse. However, only a tiny fraction of the content is meaningful and relevant. SM data can serve as a valuable channel for seeking help, offering assistance, situational awareness, general opinions and coordinating activities in disaster. At the system level, understanding the needs, availabilities, situational updates and public views enhances the planning and rescue operations, improving disaster resilience.
Timely determination in DM is vital, but it still is challenging. Deep learning (DL) algorithms are an important part of DM systems using SM data (Caragea et al., 2016; Huang et al., 2020a; Nguyen et al., 2019). Regarding the application of SM data for DM, research suggests that the lack of tools for managing SM data during a disaster makes it difficult for disaster professionals, not making them to understand how SM data can be helpful for the public. To fill this gap in the literature, this paper explores disaster data using DL techniques to determine the nature of a SM message. Our key original contributions are as follows:
A new SM-based COVID-19 data set with the label of nature of message was developed from April 22, 2021, to May 05, 2021, with 1,03,839 tweets in total.
A new fusion model was proposed for determining the nature of a disaster-related SM message by integrating the structure of CNN and bidirectional long short-term memory network (BiLSTM).
The proposed fusion model is benchmarked against other state-of-the-art models and experimental results of previous research studies.
We demonstrate the technical efficacy by determining the nature of SM messages on the COVID-19...