PAG-XIII  Plant & Animal Genomes XIII Conference

January 15-19, 2005
Town & Country Convention Center
San Diego, CA



P874 : Poster and Demo


Emperical Evaluation Of PathBinderH

LaRon M. Hughes

  Iowa State University, 3108 Coover Hall, Ames, IA, 50010, USA

PathBinderH is a Web-served text mining tool that allows users to search PubMed (including MEDLINE) to identify abstracts that contain user-specified terms co-occurring in the same sentence. Unlike standard tools that allow users to identify scientific abstracts containing one or more query terms, PathBinderH allows abstracts to be included or excluded from a search based on given plant taxa. This enables (1) filtering out abstracts dealing with species of less interest while retrieving sentences from abstracts about any of the potentially many species within the specified taxa, and (2) identifying abstracts that are more likely to prove relevant to a user than abstracts that contain the query terms but in different sentences, because the query terms are more likely to be coordinated in their use.
Preliminary results have shown that for a given query the results of a PubMed and PathBinderH query render a perhaps surprisingly small overlap in returned items. We present an empirical investigation of this important problem. We show that there is a large and significant difference in the query results of the two tools. Furthermore, the results show that PathBinderH returns results that are more relevant to researchers due to its taxonomic and sentence-based retrieval methods.