In TREC Genomics a question/answering task has been proposed. A set of questions with a specific entity of interest is proposed and a set of passages from a collection of full text documents has to be selected from the document collection provided. We have used a two step approach: the first one is recall-oriented retrieval, and the second is an information extraction system that is intended to provide higher precision. We rely on well known techniques like query expansion and resources like MeSH and UMLS. The information extraction techniques are part of the infrastructure of the Text Mining Group at European Bioinformatics Institute. Using standard information retrieval techniques has been found more beneficial than using more complex processing. Having analyzed the results we find that the performance of query expansion varies for different topics. There are several reasons. Terminological resources may contain ambiguous synonyms or synonyms whose textual usage patterns differ from the usage of the original query terms. On the whole our performance was similar to the mean results from the three performance measures.
[1]
Patrick Ruch,et al.
Vocabulary-Driven Passage Retrieval for Question-Answering in Genomics
,
2007,
TREC.
[2]
Marti A. Hearst,et al.
A Simple Algorithm for Identifying Abbreviation Definitions in Biomedical Text
,
2002,
Pacific Symposium on Biocomputing.
[3]
Dietrich Rebholz-Schuhmann,et al.
Annotation and Disambiguation of Semantic Types in Biomedical Text: A Cascaded Approach to Named Entity Recognition
,
2006,
NLPXML@EACL.
[4]
Patrick Ruch,et al.
Report on the TREC 2004 Experiment: Genomics Track
,
2004,
TREC.
[5]
Chris Buckley,et al.
Pivoted Document Length Normalization
,
1996,
SIGIR Forum.
[6]
Ellen M. Voorhees,et al.
Query expansion using lexical-semantic relations
,
1994,
SIGIR '94.