Trends in COVID-19 Publications: Streamlining Research Using NLP and LDA

Area of Science:

• Computational Biology
• Bibliometrics
• Public Health Informatics

Background:

• The rapid increase in COVID-19 research presents challenges for literature hubs in identifying complex research topics.
• Existing artificial intelligence tools struggle to capture the nuanced landscape of COVID-19 scientific discourse.
• There is a need for advanced methodologies to analyze and visualize the temporal evolution of research themes.

Purpose of the Study:

• To develop and validate a comprehensive Latent Dirichlet Allocation (LDA) model for analyzing COVID-19 research topics.
• To propose a novel methodology for visualizing temporal trends in scientific literature.
• To enhance existing online literature hubs by improving topic categorization and depth.

Main Methods:

• Utilized natural language processing (NLP) techniques to develop a 25-topic LDA model.
• Applied the model to a corpus of PubMed research articles related to "COVID."
• Integrated the methodology with LitCovid, a literature hub from the National Center for Biotechnology Information, to refine its categories.

Main Results:

• Identified significant temporal trends in COVID-19 research, including increased focus on "Mental Health" and "Socioeconomic Impact."
• Observed a decrease in research prominence for "Genome Sequence" and a stable trend for "Epidemiology."
• Revealed a research bias towards clinical applications for "masks" and "Personal Protective Equipment (PPE)," with a notable lack of population-based epidemiological studies.

Conclusions:

• The developed LDA model and visualization methodology offer a powerful tool for understanding the evolving landscape of COVID-19 research.
• The findings underscore the need for more population-based epidemiological research concerning personal protective equipment.
• The methodology can significantly improve the comprehensiveness and utility of scientific literature hubs.