KNOWLEDGE MANAGEMENT OF SOCIAL MEDIA DATA FOR DISASTER MANAGEMENT

Abstract

Disasters affect the lives, and infrastructure in a negative manner. With the internet fad, social media has become inevitable in our lives, generating tsunami of data. An individual cannot retrieve the same message on social media within a blink of eye. During disastrous times, data is even more critical. People use social media at catastrophic times. Data, being raw, and unstructured, needs to be in a knowledgable format, so that effective decisions can be taken at right time. The overarching aim of this dissertation is to advance the management of knowledge for disaster management using social media data. Conventional knowledge management systems are not optimal enough to support disaster management processes. The dynamic nature of disasters offer different situations. With the evolving times, the knowledge management systems needs to be evolved to handle complex environments. In the first objective of our research work, we explore the challenges and enablers of social media usage for disaster management by understanding views and perspectives of people working in disaster management domain. This was done with the help of literature review and data collected through focus group discussion. The participants chosen for focus group discussion are homogeneously working in disaster management domain but are from heterogeneous backgrounds like civil, architecture, mechanical, management and computer science. The number of participants are 10, ranging from 21 to 42 years of age. 8 male participants and 2 female participants are there. Half of the participants are Master students pursuing disaster management. The remaining four participants are doctoral students and one professor of Centre of Excellence in Disaster Management is amongst the 10 participants. The methodology developed is in concoction of existing literature and the efficacy of qualitative data obtained from focus group discussion, using Atlas.ti software following inductive thematic approach. The transcripts are transcribed manually by the moderator. After acquiring validation from the participants, raw data is categorized using inductive thematic approach in Atlas.ti software. The results are finalized after expert validation. The themes are developed using the panoply of coding functions - Open coding, Quick coding, List coding and In-vivo coding - available in Atlas.ti 8. The identified challenges are physical, software, cultural, demographic, authenticity, and regulatory. The identified enablers are rise in mobile penetration, democratic participation, increase in living standards, two-way real-time communication, global reach, and cheaper source of information. The study results in contribution by explaining ”what” challenges and enables the usage of social media for disaster management. The research objective sheds a new light on the understanding of social media as a vital player in disaster management and contributes to enlarge the scope of advance research on the relationship between social media and disaster management. For the second objective, The researcher wants to develop a framwork for reliable and accurate identification of disaster-related social media messages for effective management of disasters. This goal can be achieved using an efficient deep learning model with rich social media disaster-based data. In this objective, a novel deep learning based framework, iRelevancy, is proposed for identifying the disaster relevancy of a social media message using deep learning algorithms. The proposed system is evaluated with cyclone Fani data and compared with stateof- the-art deep learning models as well as the recent relevant studies. The performance of the experiments are evaluated by the accuracy, precision, recall, f1-score, area under receiver operating curve, and area under precision-recall curve score. The results show that our model is more effective for the identification of disaster-relevancy of a social media message, in comparison to other state-of-the-art methods. The predictive performance of the proposed model is illustrated with experimental results on cyclone fani data along with misclassifications. Further, to analyze the performance of the proposed model, the results on other cyclonic disasters, i.e., cyclone Titli, cyclone Amphan, and cyclone Nisarga are presented. In addition, the framework is implemented on catastrophic events of different nature, i.e., Covid-19. The research study can assist disaster managers in effectively manoeuvring disasters at the time of distress. In the third objective of our research work, we aim to propose a framework to identify the stage of disaster, i.e., pre, during, post, or irrelevant from a social media message. Extracting knowledge from the social media data during different stages of disaster management cycle is a challenging task. Deep learning has shown great potential in automatic identification from a large amount of raw data in various domains. We propose iStage, i.e., an intelligent hybrid deep learning based framework to determine the stage of the disaster to take right decisions at the right time. To demonstrate the effectiveness of iStage, it is applied on cyclonic and Covid-19 disasters. The considered disaster datasets are cyclone Fani, cyclone Titli, cyclone Amphan, cyclone Nisarga and Covid-19. The experimental results demonstrate that the proposed model outperforms Long Short-Term Memory Network and Convolutional Neural Network models. The proposed approach returns best possible solution among existing research studies considering different evaluation metrics- accuracy, precision, recall, f-score, area under receiver operating characteristic curve, and area under precision-recall curve. Hence, iStage can assist disaster personnel in a better way to manage disasters. The fourth objective aims to propose a web-based smart disaster management system for decision-making that will assist disaster professionals to determine the nature of disaster-related social media message. We consider Covid-19 as our case study. We initiate our research by re viewing the literature pertaining to the usage of social media in Covid-19, and the existing social media-based disaster management systems. Hence, we identify that the existing systems lack the web-interface, considering the importance of the social media message, due to which the most significant messages, i.e., help-seeking and help-offering are not explored. Therefore, in this objective, we address this issue with a web-application. It is worth mentioning that a fusion-based deep learning model is introduced to objectively determine the nature of a social media message. The developed system leads to a better performance in accuracy, precision, recall, F-score, area under receiver operating curve, and area under precision-recall curve, compared to other state-of-the-art methods in the literature. The contribution of this objective is three folds: Firstly, it presents a new covid dataset of social media messages with the label of nature of message. Secondly, it presents a fusion-based deep learning model to classify the social media data. Thirdly, it presents a web-based interface to visualise the structured information. Furthermore, the architecture of DisDSS is analyzed based on practical case study, i.e., covid-19. The proposed deep learning based model is embedded into a web-based interface for decision-support. To the best of our knowledge, this is the first social media-based disaster management system in India. The objectives of the dissertation deals with knowledge management of social media data for disaster management domain. The objectives transform the raw and unstructured data into a knowledgable format, so that disaster managers are able to handle disasters effectively and efficiently. The dissertation advances the academic knowledge in better understanding the role of social media for disaster management. The study is important as it helps in determining the challenges and enablers of social media usage for disaster management. The research has provided another dimension to the social media usage understanding for disaster management. As such, the study extends the inadequate knowledge of barriers and enablers. No doubt, the usage of social media during disasters is ever increasing and additional knowledge would assist in the formulation of effective policies in shielding the society from the menance of disasters. Another contribution of this research is that challenges and enablers studies are largely focused on the Western part of the world. Contrary to this, the research considers discussion group from India, which has gained little research attention thus far. The dissertation sets up some practical implications. The research recommends the proposed deep learning approach, which outperforms the baseline models, evaluating on different paramters. Another practical contribution is the creation of a Twitter dataset of cyclone Fani and Covid-19. Cyclone Fani dataset has relevancy label and stage of disaster label. Covid-19 dataset has label of nature of the social media message. Also, the results demonstrate the prediction results and web-interface.