← smac.pub home

PACRR Gated Expansion for TREC CAR 2018

pdf bibtex slides dblp: conf/trec/MacAvaneyGFY18 system description paper non-refereed

Authors: Sean MacAvaney, Andrew Yates, Nazli Goharian, Ophir Frieder

Appeared in: Proceedings of the 27th Text REtrieval Conference (TREC 2018)


In this work, we present our approach to the 2018 TREC Complex Answer Retrieval (CAR) task. We submitted two passage retrieval runs. The first uses the state-of-the-art technique from TREC CAR 2017: a modified neural ranker modified to incorporate query heading frequency information while performing term matching on each heading independently. The second run incorporates a novel gated technique for incorporating query expansion terms in a neural ranker. Our TREC runs indicate significant performance improvements can be achieved when using the expansion approach.

BibTeX @inproceedings{macavaney:trec2018-car, author = {MacAvaney, Sean and Yates, Andrew and Goharian, Nazli and Frieder, Ophir}, title = {PACRR Gated Expansion for TREC CAR 2018}, booktitle = {Proceedings of the 27th Text REtrieval Conference}, year = {2018} }