Crisis Management Using Active Online Learning Approach with Social Media Content

Abstract

Social media is used to document and share events, such as catastrophes, people encounter. It is important to utilise SM information to help crisis management by, for example, providing information about the crises that is relevant and unknown in real time. To address this problem, we present a new active online multiple-prototype classifier, which we name AOMPC. It helps find crisis-relevant data. AOMPC is an online learning method that use active learning to look for the labels of ambiguous unlabeled data in data streams. An allotted budget limits the amount of inquiries. AOMPC is often used to receive partially tagged data streams. Using data from both artificial and social media, AOMPC was evaluated for its ability to handle two crises: the Colorado floods and the Australian bushfires. A full assessment was performed to measure the quality of the outcomes, using a variety of existing tools. In addition, a sensitivity analysis was performed to reveal the influence of AOMPC's parameters on the correctness of the findings. To compare AOMPC to other existing online learning algorithms, a research was conducted. The research proved that AOMPC has excellent response to fluctuating, partly-labeled data streams.