Investigating a Key COVID-19 Question by Using Natural Language Processing on Scientific Publications

Abstract

The COVID-19 pandemic has brought an unprecedented challenge to public health. Numerous scientific publications are published daily on COVID-19 to understand the unexplored facets of the disease. The sheer volume of these publications makes it daunting for researchers to quickly find information and evaluate data related to specific COVID-19 queries. Natural Language Processing (NLP), a form of artificial intelligence, assists in churning these huge piles of data with a sophisticated algorithmic approach. The purpose of this study is to investigate key a COVID-19 question by using NLP on scientific publications. Using the T5 (Text-To-Text Transfer Transformer) model, we analyzed 740,000 journal abstracts for specific answers an important COVID-19 question. We performed qualitative observations, T-Tests (p-values and inferences), and accuracy metrics (Precision, Recall, and F1 score) to evaluate the models in this study. As the number of scientific publications increases, our proposed methodology provides an efficient mechanism for performing specific information retrieval for emerging questions, diseases, and related conditions, especially for underrepresented populations.