Multi-class classification of COVID-19 documents using machine learning algorithms

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Machine Learning

Background:

• Accurate and rapid identification of relevant scientific research papers is critical in biomedical research, especially during global health crises.
• Traditional text classifiers are insufficient for biomedical documents, necessitating the integration of diverse data sources like entities and bibliometric data.
• Developing effective biomedical document classification systems aids researchers and learners in categorizing and retrieving scientific literature efficiently.

Purpose of the Study:

• To investigate the influence of different information types and feature representation methods on biomedical document classification.
• To compare the effectiveness of various features, including text, entities, and bibliometric data, for classifying biomedical research papers.
• To provide practical guidelines for optimizing biomedical document classification systems.

Main Methods:

• Experiments were conducted using conventional text classification methods with features extracted from titles, abstracts, and bibliometric data.
• Data preprocessing involved cleaning, feature engineering, and multi-class classification across eleven input data variants.
• Ten machine learning algorithms were employed to analyze the performance, data efficiency, and interpretability of different models.

Main Results:

• TF-IDF (Term Frequency-Inverse Document Frequency) representations demonstrated superior performance compared to entity extraction methods.
• The abstract content alone proved sufficient for accurate document classification.
• Random Forest and Neural Network (BERT) algorithms achieved the highest performance across various document representations.

Conclusions:

• Biomedical document classification can be effectively achieved using TF-IDF features derived from abstracts, outperforming entity-based approaches.
• Random Forest and BERT models provide robust and accurate classification, offering practical solutions for managing biomedical literature.
• The study offers concrete guidelines for practitioners to enhance biomedical document classification, particularly relevant for managing research during health crises like COVID-19.