The Oregon Health & Science University submission to the TREC 2007 Genomics Track approached the entity list question answering task using a modular object oriented system framework. A system object coordinates a collection of processing objects into a pipe that constructs a set of queries, retrieves passage, and then processes those passages into a final output answer set. Using the framework we applied multiple levels of synonym expansion and a ranked series of topic queries with a range of specificities in order to retrieve all of the likely relevant passages with the most likely ranked higher. We then applied sentence pruning to the head and tail of each passage using both NLP and term-based techniques. Overall scores finished around the TREC Genomics mean for each of the four measures. Careful passage retrieval, including synonym expansion and multiple query construction, as well as sentence pruning was essential in achieving acceptable performance on this task.
|Original language||English (US)|
|Journal||NIST Special Publication|
|State||Published - 2007|
|Event||16th Text REtrieval Conference, TREC 2007 - Gaithersburg, MD, United States|
Duration: Nov 6 2007 → Nov 9 2007
ASJC Scopus subject areas