Applications of deep learning in extracting actionable information from crisis-related social media content

Date

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

We have witnessed a large number of crisis situations in recent years, from natural disasters to man-made disasters and also to deadly animal and human health crises, culminating with the ongoing Covid-19 public health crisis. Disasters can have devastating health and socio-economic impacts. Emergency response and critical resource management during crises are pivotal tasks in mitigating the impacts of such events. These tasks require time-critical and reliable information for effective implementation. During emergent crises, there is a huge influx of information from various sources, which makes the task of collecting and managing reliable information harder. Identifying key information relevant for emergency responders and policy makers from huge streams of data is an infeasible task for human to attempt. There is a clear need of a pipeline of systems that can monitor, identify and collect actionable and relevant information from incomplete and noisy sources of data. Social media has evolved into a platform for people to share their concerns, report information as eyewitnesses of events, and also call for help, especially during crisis situations. However, due to the unstructured nature of data shared in these digital media, inherent noise and potential misinformation, extraction of actionable information is a challenging task. Considering the challenges associated with modern data-driven emergency response and crisis management, deep-learning is a natural choice in making use of the large volume of unstructured data. However, deep-learning models, typically, require a large amount of annotated or labelled data, which may not always be available for an emergent crisis. This dissertation aims to address some of these issues by exploring multi-task and multimodal deep-learning approaches, combined with self-supervised representation learning. From an application point of view, this dissertation tackles two specific tasks surrounding crisis information management: firstly, the time-critical task of identifying actionable information for emergent crisis, and secondly, the task of analyzing public response to crisis events and the policies surrounding the events through social-media.

Description

Keywords

Deep learning, Crisis management, Multimodal learning, Social media

Graduation Month

December

Degree

Doctor of Philosophy

Department

Department of Computer Science

Major Professor

Doina Caragea

Date

2022

Type

Dissertation

Citation