Development of flood assistance system using social media and deep learning methods

Abstract

This thesis investigates the application of Twitter data and deep learning natural language processing (NLP) models for identifying tweets in urban areas during flood events where individuals require assistance. The research aims to leverage the real-time nature of social media to facilitate prompt and targeted aid delivery in urban flood situations. The study focuses on the development of deep learning NLP models capable of accurately classi fying flood-related tweets as help-seeking or non-help-seeking. By harnessing the power of these models, the thesis aims to enable efficient rescue operations by identifying and prioritizing areas and individuals in urban settings that are most in need of assistance. Ad ditionally, geocoding techniques are utilized to extract location information from identified help-seeking tweets, aiding in the geospatial targeting of rescue efforts. The findings of the study highlight the effectiveness of combining Twitter data, deep learning NLP models, and geocoding techniques to improve response times and optimize resource allocation in urban flood situations. The research contributes to the field of disaster management by show casing the potential of social media analytics and advanced NLP techniques in enhancing flood response strategies in urban environments.