License: Creative Commons Attribution 4.0 International license (CC BY 4.0)
When quoting this document, please refer to the following
DOI: 10.4230/DagSemProc.08131.3
URN: urn:nbn:de:0030-drops-15086
URL: http://dagstuhl.sunsite.rwth-aachen.de/volltexte/2008/1508/
Go to the corresponding Portal |
Briscoe, Ted ;
Gasperin, Caroline ;
Lewin, Ian ;
Vlachos, Andreas
Bootstrapping an interactive information extraction system for FlyBase curation
Abstract
We describe an adaptive information extraction (IE) system designed to
aid the curation of papers about fruit fly genomics for incorporation
into FlyBase. FlyBase employs a team of about eight curators who fill
in prespecified IE templetes (called proformas) for each gene and
allele discussed in a given paper with curatable information
associated with it. The normal approach to curation is to load the PDF
of the paper into a tool such as Acroread and to use the `Find'
function to search for repeated mentions of an entity of interest. The
relevant information is then typed into the appropriate template
fields. Templates are then checked for consistency and automatically
integrated into the database.
We have developed PaperBrowser, a tool designed to make it easier for
curators to locate relevant information. The tool takes the PDF
version of the paper as input and rerenders it as SciXML, a standard
developed at Cambridge for representing the logical structure of
scientific articles in a fashion amenable to text mining. The basic
SciXML is augmented by a gene name recogniser and anaphora
resolution module so that PaperBrowser is able to highlight gene names
in the paper and to provide a navigation bar which allows the curator
to jump to specific mentions of a given gene in the various sections
of the paper. Alternatively, the curator can select a specific gene
mention and the browser will highlight all the noun phrases which are
anaphorically linked to that gene mention. These anaphoric links can
either be coreferential, or associative to the gene's products or
components, such as proteins or RNA.
User-based evaluation of PaperBrowser in comparison to the use of
Acroread, with FlyBase curators undertaking the task of finding the
set of genes and alleles for which templates should be constructed,
has demonstrated that curation is 20\% faster at no cost to accuracy
when using PaperBrowser. PaperBrowser uses a conditional random field
model to perform gene name recognition bootstrapped from training data
derived automatically via information in FlyBase. The anaphora
resolution algorithm is unsupervised but uses information from the
Sequence Ontology augmented with lexemes from UMLS to identify noun
phrases referring to gene products and components. The PDF extraction
tool uses a commercial OCR package augmented with a seed-based machine
learning technique to learn the mapping from font and format
information to the logical structure of the paper. Papers describing
the complete processing pipeline, intrinsic evaluation of the
individual components and user-based experiments, along with test
datasets are available from the FlySlip Project website
BibTeX - Entry
@InProceedings{briscoe_et_al:DagSemProc.08131.3,
author = {Briscoe, Ted and Gasperin, Caroline and Lewin, Ian and Vlachos, Andreas},
title = {{Bootstrapping an interactive information extraction system for FlyBase curation}},
booktitle = {Ontologies and Text Mining for Life Sciences : Current Status and Future Perspectives},
pages = {1--1},
series = {Dagstuhl Seminar Proceedings (DagSemProc)},
ISSN = {1862-4405},
year = {2008},
volume = {8131},
editor = {Michael Ashburner and Ulf Leser and Dietrich Rebholz-Schuhmann},
publisher = {Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik},
address = {Dagstuhl, Germany},
URL = {https://drops.dagstuhl.de/opus/volltexte/2008/1508},
URN = {urn:nbn:de:0030-drops-15086},
doi = {10.4230/DagSemProc.08131.3},
annote = {Keywords: Biomedical Text Mining, Interactive Information Extraction, Natural Language Processing}
}
Keywords: |
|
Biomedical Text Mining, Interactive Information Extraction, Natural Language Processing |
Collection: |
|
08131 - Ontologies and Text Mining for Life Sciences : Current Status and Future Perspectives |
Issue Date: |
|
2008 |
Date of publication: |
|
03.06.2008 |