Semi-automatic indexing of full text biomedical articles

Area of Science:

• Medical Informatics
• Biomedical Text Mining
• Information Retrieval

Background:

• The U.S. National Library of Medicine's Medical Text Indexer (MTI) currently uses article titles and abstracts for indexing recommendations.
• Enhancing the MTI's input could improve the accuracy and comprehensiveness of indexing for biomedical literature.

Purpose of the Study:

• To evaluate the effectiveness of extending the Medical Text Indexer (MTI) to process the full text of medical journal articles.
• To determine which sections of a full-text article contribute most to improved indexing recommendations.

Main Methods:

• A collection of 17 journal issues comprising 500 articles was used to test the extended MTI.
• The study compared indexing performance using titles and abstracts alone versus incorporating full-text content, including various sections.
• Different models were evaluated, incorporating terms from the whole article, specific sections, titles, abstracts, captions, and untitled sections.

Main Results:

• The best performing model included the Results, Results and Discussion, and Conclusions sections, along with titles, abstracts, table/figure captions, and untitled sections.
• This comprehensive model achieved a 7.4% improvement in indexing effectiveness compared to using only titles and abstracts.

Conclusions:

• Incorporating full-text content, particularly specific sections, significantly enhances the performance of the Medical Text Indexer (MTI).
• The developed model offers a more effective approach to biomedical literature indexing, improving upon current methods.