Un chatbot para asistir a las necesidades de información en tiempos de COVID-19

TLDR: In this paper, the authors describe the design, development, and testing of a question-answering chatbot to inform COVID-19 at Cali town from March to December 2020.

Abstract: This paper aims to describe the design, development, and testing of a question-answering chatbot to inform COVID-19 at Cali town. The chatbot is based on the model of natural language processing, and it is capable of holding a question-and-answer conversation about the pandemic. This document presents the sources of information to solve information needs in the Cali town's risk scenery from March to December 2020; Based on the sources of information, a corpus with 636 sentences was built. Three models were trained bases on the corpus. The models were trained incremental prototyping: A baseline model that responds to general questions, cases, active cases, and deaths at a geographic point of an area or region of interest (BC), the baseline model, zones and news, decrees or regulations generated by the Government during the risk situation (BCN) and the final model that responds the previous items and to frequently asked questions (BFAQ). A satisfaction survey of the prototype was developed to evaluate the chatbot, and the models were evaluated by metrics of PLN precision, coverage, and F-measure. The analysis and results showed that the final model (BFAQS) showed values upper 88% in the three measurements., besides, the BFACS held 1480 conversations, with an average conversation engagement of 4.12 minutes. Furthermore, the survey results show that 87% would use the chatbot again to obtain information about COVID-19.