An Unsupervised COVID-19 Report Summarizer for Developing Smart Healthcare System

Abstract

Due to the coronavirus pandemic, COVID-19 reports are being generated frequently and volume is increasing exponentially. These reports provide innovative information to healthcare and medical science. Analysis of this information helps us to extract meaningful knowledge for improving our healthcare systems. An automated report summarizer is a model whose task is to rewrite a report into a very compact form without losing any of its important content. In the paper, the COVID-19 reports collected from the new release open-source, CORD-19 dataset, are applied in our proposed automated text summarizer. Initially, we have preprocessed the reports by Natural Language Processing (NLP) and subsequently, the Unified Medical Language System (UMLS) tool has been used to bring out only the medical terms along with the concepts corresponding to that medical terms. So, a set of concepts represent each sentence in a report and with the help of a word-to-vector-based neural network model that extracted concepts are converted into fixed dimension Word2vec vectors. Next, the fixed dimension real-valued vectors replace the set of concepts corresponding to each sentence using the summation scheme. Thus, the report is described by a real-valued dataset. Finally, the unsupervised learning (i.e. clustering algorithm) has been applied for the dataset to partition every sentence into different clusters and a representative from each cluster is identified based on Pearson correlation and clustering coefficient. The representative sentences collected using clusters are arranged according to their appearances in the report, which is the desired summary of the report. In the paper, different clustering algorithms are applied, so different A. Das (B)