Obtenção de dados estruturados a partir de diálogos para aquisição automática do domínio de planejamento gerenciador do chatbot Plantão Coronavírus

Abstract

Automated Planning is the sub-area of Artificial Intelligence (AI) that studies the deliberative process of choosing actions for an intelligent agent to achieve its goals. A planner is a general problem-solving algorithm that receives as input a high-level description of the agent and the environment (planning domain) and provides as output a sequence of actions (plan) that takes the agent from an initial configuration of the environment to a final configuration that satisfies its goals. Acquiring a planning domain is a difficult task, since it requires both knowledge of the language of actions and knowledge of the application environment. Thus, it is a process that requires intense collaboration between planning experts and experts in the application domain. This process can be done manually or automatically with algorithms that learn the description of the domain from data, called plan traces. However, obtaining this data is challenging. A viable strategy is to extract it from Natural Language information (unstructured data) that describes the process to be planned. This work aims to build a database of plan traces (structured data) from Natural Language dialogue data (unstructured data). Such data represent dialogues that occurred between real users and a chatbot called Plantão Coronavírus, which was used by the Government of the State of Ceará to assist communication between citizens and health professionals during the most critical times of the COVID-19 pandemic.