From: Comeau, Don (NIH/NLM/NCBI)
Sent: Thursday, March 03, 2005 1:28 PM
To: ncbi-seminar@ncbi.nlm.nih.gov
Subject: NCBI CBB Seminar, Tuesday, March 8 2005, 11 am

Follow Up Flag: Follow up
Flag Status: Red

NCBI CBB Seminar
Tuesday, March 8 2005, 11 am
Building 38A, Room B2N14 (NCBI Library)

Significant Phrases:  Indexing the Bookshelf

Don Comeau
NCBI/NLM/NIH

The Bookshelf is a collection of more than 30 biomedical texts
receiving nearly 600,000 hits a week.  Starting from the full text in
XML, possibly significant phrases are identified by either appearing
in titles or by repeated occurrence.  POS tagging identifies well
formed phrases.  A database preserves these results and the results of
additional human review.  Porter stemming allows recognition of
morphologically different versions of a phrase.  UMLS allows
recognition of synonymous phrases.  Synonymy offers counting and
ambiguity challenges which are addressed.  Abbreviation handling and
improved phrase determination and extraction lead efforts toward full
text understanding.