Duet at TREC 2019 Deep Learning Track

TL;DR

This paper presents three submissions using the Duet architecture for document and passage retrieval at TREC 2019, including a new multi-field model and an ensemble approach, demonstrating improved retrieval performance.

Contribution

The paper introduces DuetMF, a multi-field extension of the Duet model, and combines it with other estimators in a learning-to-rank framework, advancing neural retrieval methods.

Findings

DuetMF improves document retrieval performance.

Ensemble of Duet models enhances passage retrieval accuracy.

Combining neural and traditional estimators yields better results.

Abstract

This report discusses three submissions based on the Duet architecture to the Deep Learning track at TREC 2019. For the document retrieval task, we adapt the Duet model to ingest a "multiple field" view of documents---we refer to the new architecture as Duet with Multiple Fields (DuetMF). A second submission combines the DuetMF model with other neural and traditional relevance estimators in a learning-to-rank framework and achieves improved performance over the DuetMF baseline. For the passage retrieval task, we submit a single run based on an ensemble of eight Duet models.