Methods Inf Med 1993; 32(02): 120-130
DOI: 10.1055/s-0038-1634904
Original Article
Schattauer GmbH

Automatic Knowledge Acquisition from MEDLINE

J. J. Cimino
1   Center for Medical Informatics, Columbia University, Columbia-Presbyterian Medical Center, New York, NY, USA
,
G. O. Barnett
2   Laboratory of Computer Science, Harvard Medical School, Massachusetts General Hospital, Boston, Mass., USA
› Author Affiliations

This work was supported in part by NLM Training Grant T15-LM07037-4, UMLS Contract N01-LM-8-3513, and an equipment grant from the Hewlett-Packard Corporation. The authors wish to acknowledge Dr. Henry J. Lowe, M. D. for the data structures used to manage the MeSH terms and Ms. Laurie J. Hassan for invaluable programming assistance. The authors thank Ms. Leslie Juceam and Dr. George Hripcsak for editorial advice.
Further Information

Publication History

Publication Date:
08 February 2018 (online)

Preview

Abstract:

Construction of medical knowledge bases for use in expert systems is an arduous task. We propose a procedure for obtaining medical knowledge via automated analysis of citations found in the National Library of Medicine’s Medline ® database. In this method, simple pattern of keywords and subheading co-occurrences are detected in the keyword descriptor portion of the citations. Each pattern corresponds to a fact, expressed as a semantic relationship between medical concepts. We have constructed a set of 504 pattern-matching rules and applied it to a set of 673 Medline ® citations to produce 2,795 such facts. The results are presented of an analysis of the syntactic and semantic features of these facts to understand the kinds of knowledge than can be obtained through our method and speculate on the potential uses and pitfalls for knowledge of this type.