Related Experiment Video
Updated: Jul 9, 2026
07:35
Selecting Multiple Biomarker Subsets with Similarly Effective Binary Classification Performances
Published on: October 11, 2018
A comparison study on algorithms of detecting long forms for short forms in biomedical text
Manabu Torii1, Zhang-zhi Hu, Min Song
1Department of Biostatistics, Bioinformatics, and Biomathematics, Georgetown University Medical Center, 4000 Resevoir Rd, NW, Washington, DC 20057, USA. mt352@georgetown.edu
BMC Bioinformatics
|December 6, 2007
Summary
This study evaluates systems for detecting biomedical acronym definitions, finding that existing knowledge bases cover frequently used terms well. A web interface integrates these resources for easier access to definitions.
Area of Science:
- Biomedical informatics
- Natural Language Processing
- Literature Mining
Background:
- Growing need for literature mining systems in biomedicine.
- Focus on detecting definitions of short forms (acronyms, abbreviations, symbols) and their long forms in biomedical text.
- Evaluation of system performance, knowledge base coverage, and integration strategies for short form detection.
Purpose of the Study:
- Assess the performance of three distinct systems in identifying long forms for short forms in novel biomedical text.
- Investigate the terminological coverage of the Unified Medical Language System (UMLS) and BioThesaurus for short form synonyms.
- Develop a method to integrate results from multiple short form knowledge bases.
Main Methods:
- Evaluated three publicly available short form detection systems: Ao and Takagi's ALICE (rule-based), Chang et al.'s machine learning system, and Schwartz and Hearst's alignment-based program.
- Assessed the conceptual coverage of UMLS and BioThesaurus for short form-long form relationships.
- Implemented a web interface for virtual integration of various short form knowledge bases.
Main Results:
- Detection systems demonstrated high agreement in identifying long forms for short forms.
- Existing terminological knowledge bases exhibit good coverage of synonymous relationships for frequently defined long forms.
- The developed web interface facilitates the detection of short form definitions and searching across multiple knowledge bases.
Conclusions:
- Current systems and knowledge bases are effective for detecting definitions of common biomedical short forms.
- Integration of multiple knowledge bases through the web interface enhances the utility of biomedical literature mining.
- The findings support the development of more robust literature mining applications by leveraging existing resources and improved integration.
