Related Experiment Videos
Automated knowledge extraction from MEDLINE citations
1Department of Medical Informatics, Columbia University, New York, NY, USA.
Proceedings. AMIA Symposium
|November 18, 2000
Summary
This study explored automating knowledge base construction using MeSH term co-occurrence in medical literature. Preliminary results show significant co-occurrence patterns, suggesting potential for automated information retrieval in evidence-based medicine.
Area of Science:
- Medical Informatics
- Knowledge Representation
- Computational Linguistics
Background:
- Developing digital libraries requires efficient knowledge base construction.
- Automating knowledge extraction from biomedical literature is a key challenge.
- Evidence-based medicine relies on effective retrieval of relevant medical information.
Purpose of the Study:
- To investigate the feasibility of automating knowledge base construction.
- To identify relevant Unified Medical Language System (UMLS) semantic types for clinical questions.
- To analyze MeSH term co-occurrence in MEDLINE for knowledge discovery.
Main Methods:
- Utilized MeSH term co-occurrence within MEDLINE citations.
- Applied optimal search strategies for evidence-based medicine.
- Employed UMLS semantic types to categorize search results by question type (etiology, diagnosis, therapy, prognosis).
Main Results:
- An automated process yielded substantial information.
- Seven to eight percent of generated semantic pairs showed significant co-occurrence.
- Pilot study demonstrated good specificity and sensitivity for the project's goals.
Conclusions:
- MeSH term co-occurrence analysis is a viable method for automated knowledge base construction.
- UMLS semantic types effectively differentiate clinical question types.
- This approach shows promise for enhancing digital libraries and evidence-based medicine.