Target specific mining of COVID-19 scholarly articles using one-class approach

Area of Science:

• Computational Biology
• Data Science
• Infectious Disease Research

Background:

• The COVID-19 pandemic, caused by SARS-CoV-2, has led to a surge in research publications.
• Manually extracting relevant information from the extensive body of COVID-19 literature is impractical.
• Efficiently identifying research trends and activities is crucial for advancing prevention and treatment strategies.

Purpose of the Study:

• To develop and validate a machine learning approach for analyzing and categorizing COVID-19 research articles.
• To assist the research community in navigating the vast scientific literature on coronavirus disease 2019 (COVID-19).
• To identify trends and activities within COVID-19 research for future exploration of prevention and treatment techniques.

Main Methods:

• Utilized the COVID-19 Open Research Dataset (CORD-19) for experimental analysis.
• Employed clustering techniques, specifically k-means, to group similar research articles.
• Applied parallel one-class support vector machines (OCSVMs) for task assignment and classification of article clusters.

Main Results:

• The combination of k-means clustering followed by parallel OCSVMs demonstrated superior performance in categorizing research articles.
• The proposed method proved effective in both original and reduced feature spaces, validating its robustness.
• The approach successfully mined target-class guided information, revealing patterns in COVID-19 research.

Conclusions:

• The machine learning methodology, particularly k-means clustering with parallel OCSVMs, is an effective tool for analyzing large-scale research datasets like CORD-19.
• This approach facilitates efficient exploration of scientific literature, aiding researchers in identifying key trends and knowledge gaps in COVID-19 research.
• The findings support the use of advanced data analytics for accelerating scientific discovery in response to global health crises.