A bottom-up approach to MEDLINE indexing recommendations

Area of Science:

• Biomedical informatics
• Medical librarianship
• Natural language processing

Background:

• MEDLINE indexing is crucial for organizing biomedical literature, traditionally done manually by US National Library of Medicine staff.
• The Medical Text Indexer (MTI) program has assisted this process since 2002, aiming to improve efficiency and consistency.
• Accurate MeSH term assignment is vital for effective literature searching and retrieval.

Purpose of the Study:

• To develop and evaluate a novel bottom-up approach for automated MEDLINE indexing.
• To enhance the accuracy and efficiency of Medical Subject Headings (MeSH) recommendations generated by the Medical Text Indexer (MTI).
• To investigate the utility of supervised machine learning combined with triage rules for improving MeSH term suggestions.

Main Methods:

• A two-step process was implemented, analyzing abstracts for indicators to recommend specific MeSH terms.
• Supervised machine learning algorithms were employed to identify patterns and relationships within the text.
• Triage rules were integrated to refine recommendations and manage the number of suggested terms.

Main Results:

• The proposed approach demonstrated improved sensitivity in MeSH term recommendations.
• The combination of machine learning and triage rules effectively controlled the number of recommended terms.
• The study observed a notable enhancement in the quality of MTI recommendations.

Conclusions:

• The bottom-up approach shows promise for improving automated MEDLINE indexing.
• Further research is warranted to explore this method across a broader range of MeSH headings.
• This technique has the potential to optimize the efficiency and accuracy of biomedical literature indexing.