Construction of a disaster-support dynamic knowledge chatbot

• Autor(a) principal: Boné, João Miguel Baptista
• Data de Publicação: 2020
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10071/22066
• topic: Disaster-management; Natural language processing; Artificial intelligence; Machine learning; Chatbots; Graph databases; Gestão de desastres; Processamento da língua natural; Inteligência artificial; Aprendizagem automática; Bases de dados em grafos

This dissertation is aimed at devising a disaster-support chatbot system with the capacity to enhance citizens and first responders’ resilience in disaster scenarios, by gathering and processing information from crowd-sensing sources, and informing its users with relevant knowledge about detected disasters, and how to deal with them. This system is composed of two artifacts that interact via a mediator graph-structured knowledge base. Our first artifact is a crowd-sourced disaster-related knowledge extraction system, which uses social media as a means to exploit humans behaving as sensors. It consists in a pipeline of natural language processing (NLP) tools, and a mixture of convolutional neural networks (CNNs) and lexicon-based models for classifying and extracting disasters. It then outputs the extracted information to the knowledge graph (KG), for presenting connected insights. The second artifact, the disaster-support chatbot, uses a state-of-the-art Dual Intent Entity Transformer (DIET) architecture to classify user intents, and makes use of several dialogue policies for managing user conversations, as well as storing relevant information to be used in further dialogue turns. To generate responses, the chatbot uses local and official disaster-related knowledge, and infers the knowledge graph for dynamic knowledge extracted by the first artifact. According to the achieved results, our devised system is on par with the state-of-the- art on Disaster Extraction systems. Both artifacts have also been validated by field specialists, who have considered them to be valuable assets in disaster-management.