Machine learning for biomedical literature triage.

Area of Science:

• Bioinformatics
• Machine Learning
• Natural Language Processing

Background:

• Manual curation of biological literature is essential for knowledge extraction.
• The initial step, triage, involves classifying literature relevance, which is often challenging due to imbalanced datasets.

Purpose of the Study:

• To develop and evaluate machine learning models for automating the biological literature triage process.
• To identify the optimal classification approach for handling imbalanced datasets in literature triage.

Main Methods:

• Comparison of multiple classification models including Naive Bayes, Support Vector Machine, and Logistic Model Trees.
• Experimentation with dataset sampling techniques (under-sampling) and domain-relevant features.
• Performance evaluation of different machine learning algorithms on the triage task.

Main Results:

• The Logistic Model Trees algorithm demonstrated superior performance in classifying imbalanced datasets for literature triage.
• The combination of domain-relevant features and an under-sampling technique significantly improved model accuracy.
• Specific machine learning algorithms and feature sets were identified as critical for effective triage.

Conclusions:

• Machine learning, particularly Logistic Model Trees with appropriate features and sampling, can effectively support biological literature triage.
• The developed system offers a promising solution for improving the efficiency and accuracy of the manual curation process.
• This approach addresses the challenge of imbalanced data in automated literature classification for biological research.