Automatic resolution of ambiguous terms based on machine learning and conceptual relations in the UMLS

Area of Science:

• Biomedical Natural Language Processing
• Medical Informatics
• Computational Linguistics

Background:

• The Unified Medical Language System (UMLS) is crucial for natural language processing (NLP) in healthcare.
• Mapping free text to UMLS concepts requires disambiguating terms with multiple meanings.
• Manual annotation of corpora for disambiguation is costly and time-consuming.

Purpose of the Study:

• To develop an automated method for constructing sense-tagged corpora for ambiguous UMLS terms.
• To improve the accuracy of mapping free text to UMLS concepts.
• To reduce the manual effort required for corpus annotation.

Main Methods:

• Extracted MEDLINE abstracts containing ambiguous terms.
• Identified UMLS concept relations within abstracts.
• Automatically annotated term senses to create a sense-tagged corpus.
• Evaluated the corpus using precision and recall metrics.

Main Results:

• Achieved an overall precision of 92.9% and recall of 47.4% for the derived sense-tagged corpus.
• Refined results showed 96.8% precision and 50.6% recall after filtering rare senses and closely related abbreviations.
• Demonstrated the feasibility of automatic sense-tagging for biomedical abbreviations.

Conclusions:

• UMLS conceptual relations and MEDLINE abstracts enable automatic knowledge acquisition for disambiguation.
• This method offers an efficient alternative to manual annotation for building sense-tagged corpora.
• The approach facilitates improved free-text to UMLS concept mapping in NLP applications.