Neural Machine Reading Comprehension on COVID Dataset using Bi-directional Encoder Representations from Transformers

ABSTRACT

Global impact of COVID-19 disrupted normal life, during the pandemic enormous amount of unreliable information about COVID-19 is creating and disseminating every minute, which leads to information flooding. So people cannot extract reliable information from these unreliable sources. Machine Reading Comprehension can alleviate this problem. Machine Reading Comprehension (MRC) is a core process in question answering systems and that efficiently extracts the answer from relevant resources automatically for the questions posed by humans. Machine reading comprehension brings attention to a textual understanding with answering questions. This article introduces a new dataset, COVIDATA for the span extraction task of MRC in the domain of COVID-19 and exploits the advancements of the Transformer-BERT fine-tuned model. COVIDATA dataset generated from the COVID-19 news and information from the World Health Organization. Fine-tuned BERT, SciBERT, BioBERT, ClinicalBERT, and Bio+Clinical BERT models are used for the experiment and evaluation of the COVIDATA dataset. © 2021 IEEE.