Vocabulary-Driven Passage Retrieval for Question-Answering in Genomics

Julien Gobeill, Imad Tbahriti, Frédéric Ehrler, Patrick Ruch
2007 Text Retrieval Conference  
This year, like in 2006, we use a collection of about 160000 full-text articles. The proposed task is a passage retrieval task. Several measures are applied on the returned data: passage mean average precision, document mean average precision, aspect mean average precision. This year, our efforts concentrated on combining knowledge-driven methods on top of a standard vectorspace retrieval approach. We tested a passage selection methods based on vocabulary density estimation using several
more » ... logies of the domain. We also attempted to improve standard retrieval approaches based on vector-space similarities by using a Boolean completion principle, which overweight documents containing all keywords. These combinations did not result in a significant improvement compared to the baseline system (document map ~ 0.20) and current results do not show much improvement compared to last year's reported results.
dblp:conf/trec/GobeillTER07 fatcat:xikbtirdg5gxldyjijowzyxr3m