Information Extraction from CORD-19 Using Hierarchical Clustering and Word Bank
Rushit Jain, Bhavesh Bellaney, Parth Jangid · 2021
The Coronavirus pandemic has called for extensive research in the medical discipline. Since such disease outbreaks are about life and death of the patients, the doctors’ and biomedical scientists’ time is crucial. Research documents are usually comprehensive, often consuming the readers’ time. A solution to it would be to extract information from the research text resembling the most relevant parts of the original text so that their valuable time is saved. The problem of text summarization is to create a shortened piece of text that represents the most relevant information from a relatively larger piece of text. This paper aims to ease the burden of the doctors so won't have to read the extensive research documents by constructing a summary of the most relevant parts of a medical research paper. A text summarization algorithm always works by quantifying the sentences by some means and analyzing scores. We use the TF-IDF quantification which is a popular way to quantify sentences. We select sentences with a high score and exclude those with a lower score, compared to a threshold. In a medical research paper, several sentences might have a low score, but they might be important if they contain biomedical entities. We use a dataset which has been constructed upon biomedical and COVID-19 terminology to construct a much better summary than some existing tools. As new methods keep coming up, this simple, yet robust approach gives us an accuracy of over 78% for most CORD-19 research papers.