An Analysis of BERT FAQ Retrieval Models for COVID-19 Infobot

Shuo Sun, João Sedoc

Keywords: COVID-19, BERT, QA retrieval, infobot

Abstract: The outbreak of the COVID-19 pandemic has caused tremendous amounts of suffering and deaths around the world and greatly affected the lives of humanity. As the world sees more infected cases every day, the need and demand for reliable and up-to-date information on COVID-19 have never been higher. While recent pre-trained language models show successes on many other NLP tasks, we did not have COVID-19 related dataset to help us evaluate the performance of QA systems and infobots based on these models. After the creation of a COVID-19 question similarity dataset by public health experts from the Johns Hopkins Bloomberg School of Public Health (JHSPH), we create models sufficient for application. We also analyze the amount of supervised data required.