Indexing UMLS Semantic Types for Medical Question-Answering

Area of Science:

• Medical Informatics
• Natural Language Processing
• Biomedical Text Mining

Background:

• Open-domain Question-Answering (QA) systems commonly use general named entities.
• Medical QA systems require specialized, medically relevant named entities.
• The Unified Medical Language System (UMLS) Semantic Network offers a rich source of medical semantic types.

Purpose of the Study:

• To explore the utility of UMLS Semantic Network semantic types for medical QA.
• To develop and evaluate a medically-specific named entity tagger using French UMLS resources.
• To investigate the detection of UMLS Semantic Network relations for medical question answering.

Main Methods:

• Utilized the French part of the UMLS Metathesaurus and its semantic types.
• Developed a medically-specific named entity tagger.
• Explored Semantic Network relation detection for question answering.
• Evaluated performance on a French-language medical document corpus from the EQueR forum.

Main Results:

• Demonstrated the effectiveness of UMLS semantic types for medical named entity recognition.
• Showcased the capability of detecting Semantic Network relations for medical QA.
• Statistical studies indicated the importance of considering document origin in QA strategies.

Conclusions:

• UMLS Semantic Network semantic types are valuable resources for building specialized medical QA systems.
• Integrating document origin into QA strategies enhances performance for medical text.
• The developed tagger and relation detection methods show promise for medical information retrieval.