Machine Learning-Based Approach for Identifying Research Gaps: COVID-19 as a Case Study.

Area of Science:

• Computational Biology
• Bibliometrics
• Artificial Intelligence in Medicine

Background:

• Traditional methods for identifying research gaps are time-consuming and potentially biased.
• Scalable and innovative approaches are needed to systematically assess scientific literature.
• The COVID-19 pandemic highlighted the need for efficient literature analysis.

Purpose of the Study:

• To propose and evaluate a machine learning-based approach for identifying research gaps.
• To utilize the COVID-19 Open Research dataset for a case study.
• To demonstrate the potential of automated methods in scientific discovery.

Main Methods:

• Employed the BERTopic technique for topic modeling on the CORD-19 dataset.
• Utilized transformer models and class-based TF-IDF for document embedding and clustering.
• A three-stage process involved document embedding, clustering, and topic representation.

Main Results:

• Identified 21 distinct research gap areas within the COVID-19 literature.
• Grouped these gaps into six principal topics: virus, risk factors, prevention, treatment, healthcare delivery, and impact.
• The 'impact of COVID-19' was the most prominent topic, appearing in over half the studies.

Conclusions:

• The machine learning approach effectively identifies research gaps, serving as a guide for future research queries.
• This method complements, rather than replaces, traditional literature reviews.
• Future work should incorporate updated literature, full-text analysis, and advanced modeling techniques.