RECORD: a Question Answering tool for COVID-19

Matteo Muffo, Aldo Cocco, Mattia Messina, Enrico Bertino

Keywords: COVID-19, SARS-CoV-2, Information Retrieval, Question Answering

TL;DR: Question Answering tool to answer COVID-related questions

Abstract: In this work we present RECORD (Research Engine for COVID Open Research Dataset), a tool to provide answers to questions related to COVID-19. The engine that we propose adopts a two-step pipeline to extract answers from the CORD-19 dataset, the most complete collection of scientific articles about COVID-19 disease and SARS-CoV-2. RECORD leverages a large pre-trained language model to select, from a given input question, the most semantically-related papers, then a Question Answering model is used to provide the best answer for each selected paper. In order to evaluate the performances of our tool, we tested it on two sets of questions (one proposed by our evaluation team plus one proposed by the health facility Centro Medico Sant’Agostino), adopting a scoring mechanism to assess the quality of the outcomes. In this context, RECORD has shown remarkable results in terms of consistency and precision of the extracted answers.