Geração de dados sintéticos para treinamento de chatbots baseados na carta de serviços do Governo do Ceará

Abstract

The list of services of the Government of Ceará is a set of information about the services offered by the Government. Finding ways to facilitate access to this information is one of the objectives of the Digital Government project. Considering this objective, the present work is an analysis of methods to generate databases for a domain not yet explored, using the list of services as a domain. Using data generation techniques based on text models and automatic text generation using the PTT5 deep learning model, databases are generated considering the application in these databases in a chatbot developed in Rasa, focusing on the task classification of intentions. The classification task is performed through models based on transformers, exploring more restricted and less restricted cases. The results obtained using the chatbot and the intention classification models developed in this work are promising and show that the data generation techniques can be well explored in order to generate a quality database.