| Abstract
|
Conventional methods for retrieving inform … Conventional methods for retrieving information from the medical literatureare imprecise and inefficient. Information retrieval systems employunmanageable indexing vocabularies or use full-text representations thatoverwhelm the user with irrelevant information. This paper describes adocument representation designed to improve the precision of searching intextual databases without significantly compromising recall. Therepresentation augments simple text word representations with contextualmodels that reflect recurring semantic themes in clinical publications. Using this representation, a searcher may indicate both the terms ofinterest and the contexts in which they should occur. The contexts limitthe potential interpretations of text words, and thus form the basis formore precise searching. In this paper, we discuss the shortcomings oftraditional retrieval systems and describe our context-basedrepresentation. Improved retrieval performance with contextual models isillustrated by example, and a more extensive study is proposed. We presentan evaluation of the contextual models as an indexing scheme, using avariation of the traditional inter-indexer consistency experiments, and wedemonstrate that contextual indexing is reproducible by minimally trainedphysicians and medical students. ly trainedphysicians and medical students.
|
| Author
|
Gretchen P. Purcell +,
Edward H. Shortliffe +
|
| Bibtype
|
techreport +
|
| Institution
|
Knowledge Systems, AI Laboratory +
|
| Key
|
KSL-95-48 +
|
| Modification dateThis property is a special property in this wiki.
|
1 May 2009 14:06:08 +
|
| Month
|
May +
|
| Note
|
Updated September 1995. Medical Computer Science +
|
| Number
|
KSL-95-48 +
|
| Tag
|
Computer science +
|
| Title
|
Contextual Models of Clinical Publications for Enhancing Retrieval from Full-Text Databases +
|
| Tr id
|
KSL-95-48 +
|
| Year
|
1995 +
|
| Categories |
Technical Report,
Publication,
KSL Technical Report
|
