Creating an online dictionary of abbreviations from MEDLINE

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Computational Biology

Background:

• The rapid expansion of biomedical literature is hindered by the uncontrolled use of abbreviations.
• Abbreviations increase the effective vocabulary size, posing challenges for human readers and automated systems.
• Developing an automated method to identify and expand abbreviations is crucial for information retrieval.

Purpose of the Study:

• To develop and evaluate an algorithm for automatically matching abbreviations with their expansions in biomedical texts.
• To create a comprehensive lexicon of biomedical abbreviations.

Main Methods:

• A logistic regression model was employed to score potential abbreviation expansions based on a human-annotated training set.
• The algorithm was applied to the Medstract corpus and subsequently to all MEDLINE abstracts.
• Performance was evaluated using recall and precision metrics on independent datasets.

Main Results:

• The algorithm achieved up to 83% recall at 80% precision on the Medstract corpus.
• A database of 781,632 high-scoring abbreviations was generated from MEDLINE.
• 88% of abbreviations from an independent list were successfully identified in the database.

Conclusions:

• An effective algorithm for identifying biomedical abbreviations from text has been developed.
• The algorithm and generated abbreviation database are publicly accessible via a web server.
• This work facilitates better information extraction and management within the biomedical domain.