When asked, large language models (LLMs) like ChatGPT claim that they can assist with relevance judgments but it is not clear whether automated judgments can reliably be used in evaluations of retrieval systems. In this perspectives paper, we discuss possible ways for LLMs to support relevance judgments along with concerns and issues that arise. We devise a human--machine collaboration spectrum that allows to categorize different relevance judgment strategies, based on how much humans rely on machines. For the extreme point of ``fully automated judgments'', we further include a pilot experiment on whether LLM-based relevance judgments correlate with judgments from trained human assessors. We conclude the paper by providing opposing perspectives for and against the use of~LLMs for automatic relevance judgments, and a compromise perspective, informed by our analyses of the literature, our preliminary experimental evidence, and our experience as IR researchers.
%0 Journal Article
%1 Faggioli2023-sl
%A Faggioli, Guglielmo
%A Dietz, Laura
%A Clarke, Charles
%A Demartini, Gianluca
%A Hagen, Matthias
%A Hauff, Claudia
%A Kando, Noriko
%A Kanoulas, Evangelos
%A Potthast, Martin
%A Stein, Benno
%A Wachsmuth, Henning
%D 2023
%I arXiv
%K topic_language
%T Perspectives on large language models for relevance judgment
%X When asked, large language models (LLMs) like ChatGPT claim that they can assist with relevance judgments but it is not clear whether automated judgments can reliably be used in evaluations of retrieval systems. In this perspectives paper, we discuss possible ways for LLMs to support relevance judgments along with concerns and issues that arise. We devise a human--machine collaboration spectrum that allows to categorize different relevance judgment strategies, based on how much humans rely on machines. For the extreme point of ``fully automated judgments'', we further include a pilot experiment on whether LLM-based relevance judgments correlate with judgments from trained human assessors. We conclude the paper by providing opposing perspectives for and against the use of~LLMs for automatic relevance judgments, and a compromise perspective, informed by our analyses of the literature, our preliminary experimental evidence, and our experience as IR researchers.
@article{Faggioli2023-sl,
abstract = {When asked, large language models (LLMs) like ChatGPT claim that they can assist with relevance judgments but it is not clear whether automated judgments can reliably be used in evaluations of retrieval systems. In this perspectives paper, we discuss possible ways for LLMs to support relevance judgments along with concerns and issues that arise. We devise a human--machine collaboration spectrum that allows to categorize different relevance judgment strategies, based on how much humans rely on machines. For the extreme point of ``fully automated judgments'', we further include a pilot experiment on whether LLM-based relevance judgments correlate with judgments from trained human assessors. We conclude the paper by providing opposing perspectives for and against the use of~LLMs for automatic relevance judgments, and a compromise perspective, informed by our analyses of the literature, our preliminary experimental evidence, and our experience as IR researchers.},
added-at = {2024-09-10T10:41:24.000+0200},
author = {Faggioli, Guglielmo and Dietz, Laura and Clarke, Charles and Demartini, Gianluca and Hagen, Matthias and Hauff, Claudia and Kando, Noriko and Kanoulas, Evangelos and Potthast, Martin and Stein, Benno and Wachsmuth, Henning},
biburl = {https://puma.scadsai.uni-leipzig.de/bibtex/2738ac3b7a31bf505b2c5d34832e26090/scadsfct},
interhash = {ee7c0af44c4c4c605f9f7449aee52dae},
intrahash = {738ac3b7a31bf505b2c5d34832e26090},
keywords = {topic_language},
publisher = {arXiv},
timestamp = {2024-11-28T17:41:18.000+0100},
title = {Perspectives on large language models for relevance judgment},
year = 2023
}
