Recommending MeSH terms for annotating biomedical articles.

Area of Science:

• Biomedical Informatics
• Information Retrieval
• Natural Language Processing

Background:

• Manual curation of scientific literature is costly and time-consuming.
• Automated methods are needed to assist in indexing biomedical literature.
• MeSH (Medical Subject Headings) indexing is crucial for archiving and retrieving MEDLINE citations.

Purpose of the Study:

• To develop a novel automated approach for MeSH indexing.
• To improve the accuracy and efficiency of assigning MeSH terms to MEDLINE citations.

Main Methods:

• The study reformulates MeSH indexing as a ranking problem.
• A learning-to-rank algorithm (ListNet) is used to rank MeSH headings.
• The algorithm utilizes neighbor documents to identify relevant MeSH terms.

Main Results:

• The proposed method achieved a precision of 0.390, recall of 0.712, and MAP of 0.626 on a benchmark dataset.
• Statistically significant improvements of up to 39% in Mean Average Precision (MAP) were observed compared to state-of-the-art methods.
• Similar significant improvements were noted on a larger dataset of 1000 documents.

Conclusions:

• The developed approach demonstrates the most accurate MeSH predictions to date.
• The method shows great potential for practical impact on MeSH indexing.
• The learning framework is robust and adaptable to other biomedical indexing tasks.