0

Declarative Experimentation in Information Retrieval using PyTerrier

A framework called PyTerrier enables efficient and declarative design, compilation, and optimization of information retrieval pipelines using deep learning concepts.

Year
2020
Venue
arXiv 2020
Authors
2
Hosting
Abstract onlyARXIV-DEFAULT

Cite

Notes

Only stored in your browser.

Attribution

Abstract & full text
arxiv.org/abs/2007.14271ARXIV-DEFAULT
TL;DR
Semantic Scholar
Attribution policy →

Abstract

The advent of deep machine learning platforms such as Tensorflow and Pytorch, developed in expressive high-level languages such as Python, have allowed more expressive representations of deep neural network architectures. We argue that such a powerful formalism is missing in information retrieval (IR), and propose a framework called PyTerrier that allows advanced retrieval pipelines to be expressed, and evaluated, in a declarative manner close to their conceptual design. Like the aforementioned frameworks that compile deep learning experiments into primitive GPU operations, our framework targets IR platforms as backends in order to execute and evaluate retrieval pipelines. Further, we can automatically optimise the retrieval pipelines to increase their efficiency to suite a particular IR platform backend. Our experiments, conducted on TREC Robust and ClueWeb09 test collections, demonstrate the efficiency benefits of these optimisations for retrieval pipelines involving both the Anserini and Terrier IR platforms.

Authors

2