University of Glasgow Terrier Team at the TREC 2021 Deep Learning Track

pdf bibtex system description paper non-refereed

Authors: Xiao Wang, Sean MacAvaney, Craig Macdonald, Iadh Ounis

Appeared in: Proceedings of the 30th Text REtrieval Conference (TREC 2021)

Abstract:

This paper describes our submission to the document ranking and passage ranking tasks of the TREC 2021 Deep Learning Track. In our participation, we conducted the dense retrieval and the sparse retrieval as well as the hybrid of dense and sparse retrieval on both passage ranking and document ranking tasks. For the dense retrieval experiments, we employed the multiple representation ColBERT dense retrieval with and without the pseudo-relevance feedback mechanism implemented. For the sparse retrieval experiments, we experimented with the sparse retrieval model, namely DPH, with and without the query expansion applied, then followed with different types of neural rerankers. For the passage ranking task, we submitted three group runs with the dense retrieval applied: uogTrPC, uogTrPCP and uogTrPot6, and two baseline runs on the inverted index: uogTrPD and uogTrPPD. For the document ranking task, we submitted three group runs with dense retrieval applied: uogTrDCPpmp, uogTrDot5pmp and uogTrDDQ5, and five sparse retrieval baseline runs. For both tasks, after correction, the hybrid of sparse and dense retrieval runs, namely uogTrPot6-c run for passage ranking task and uogTrDot5pmp-c run for document ranking task, are most effective.

BibTeX @inproceedings{wang:trec2021-dl, author = {Wang, Xiao and MacAvaney, Sean and Macdonald, Craig and Ounis, Iadh}, title = {University of Glasgow Terrier Team at the TREC 2021 Deep Learning Track}, booktitle = {Proceedings of the 30th Text REtrieval Conference}, year = {2021} }