Is ChatGPT Good at Search? Investigating Large Language Models as Re-Ranking Agents

Large Language Models (LLMs) have demonstrated remarkable zero-shot generalization across various language-related tasks, including search engines. However, existing work utilizes the generative ability of LLMs for Information Retrieval (IR) rather than direct passage ranking. The discrepancy betwee...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Sun, Weiwei, Yan, Lingyong, Ma, Xinyu, Wang, Shuaiqiang, Ren, Pengjie, Chen, Zhumin, Yin, Dawei, Ren, Zhaochun
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page
container_issue
container_start_page
container_title
container_volume
creator Sun, Weiwei
Yan, Lingyong
Ma, Xinyu
Wang, Shuaiqiang
Ren, Pengjie
Chen, Zhumin
Yin, Dawei
Ren, Zhaochun
description Large Language Models (LLMs) have demonstrated remarkable zero-shot generalization across various language-related tasks, including search engines. However, existing work utilizes the generative ability of LLMs for Information Retrieval (IR) rather than direct passage ranking. The discrepancy between the pre-training objectives of LLMs and the ranking objective poses another challenge. In this paper, we first investigate generative LLMs such as ChatGPT and GPT-4 for relevance ranking in IR. Surprisingly, our experiments reveal that properly instructed LLMs can deliver competitive, even superior results to state-of-the-art supervised methods on popular IR benchmarks. Furthermore, to address concerns about data contamination of LLMs, we collect a new test set called NovelEval, based on the latest knowledge and aiming to verify the model's ability to rank unknown knowledge. Finally, to improve efficiency in real-world applications, we delve into the potential for distilling the ranking capabilities of ChatGPT into small specialized models using a permutation distillation scheme. Our evaluation results turn out that a distilled 440M model outperforms a 3B supervised model on the BEIR benchmark. The code to reproduce our results is available at www.github.com/sunnweiwei/RankGPT.
doi_str_mv 10.48550/arxiv.2304.09542
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2304_09542</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2304_09542</sourcerecordid><originalsourceid>FETCH-LOGICAL-a672-5c050871a953f3e73310c1c5b0eea4651ec7f9b7d8b65c2cf1d75e47bd34bbb13</originalsourceid><addsrcrecordid>eNotj0FOwzAURL1hgQoHYIUvkGDHdpysUBVBiJQK1GYffTs_bkRxkB0quD1tYTMzi6eRHiF3nKWyUIo9QPiejmkmmExZqWR2TXZNpNUelvqto_U8DxQWukMIdv9IG3_EuEwOlsk72kJweErvvuA0NvOAh0gh0i0mW_DvZ2bt0C_xhlyNcIh4-98r0j0_ddVL0r7WTbVuE8h1lijLFCs0h1KJUaAWgjPLrTIMEWSuOFo9lkYPhcmVzezIB61QajMIaYzhYkXu_24vVv1nmD4g_PRnu_5iJ34BDx5JVA</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>Is ChatGPT Good at Search? Investigating Large Language Models as Re-Ranking Agents</title><source>arXiv.org</source><creator>Sun, Weiwei ; Yan, Lingyong ; Ma, Xinyu ; Wang, Shuaiqiang ; Ren, Pengjie ; Chen, Zhumin ; Yin, Dawei ; Ren, Zhaochun</creator><creatorcontrib>Sun, Weiwei ; Yan, Lingyong ; Ma, Xinyu ; Wang, Shuaiqiang ; Ren, Pengjie ; Chen, Zhumin ; Yin, Dawei ; Ren, Zhaochun</creatorcontrib><description>Large Language Models (LLMs) have demonstrated remarkable zero-shot generalization across various language-related tasks, including search engines. However, existing work utilizes the generative ability of LLMs for Information Retrieval (IR) rather than direct passage ranking. The discrepancy between the pre-training objectives of LLMs and the ranking objective poses another challenge. In this paper, we first investigate generative LLMs such as ChatGPT and GPT-4 for relevance ranking in IR. Surprisingly, our experiments reveal that properly instructed LLMs can deliver competitive, even superior results to state-of-the-art supervised methods on popular IR benchmarks. Furthermore, to address concerns about data contamination of LLMs, we collect a new test set called NovelEval, based on the latest knowledge and aiming to verify the model's ability to rank unknown knowledge. Finally, to improve efficiency in real-world applications, we delve into the potential for distilling the ranking capabilities of ChatGPT into small specialized models using a permutation distillation scheme. Our evaluation results turn out that a distilled 440M model outperforms a 3B supervised model on the BEIR benchmark. The code to reproduce our results is available at www.github.com/sunnweiwei/RankGPT.</description><identifier>DOI: 10.48550/arxiv.2304.09542</identifier><language>eng</language><subject>Computer Science - Computation and Language ; Computer Science - Information Retrieval</subject><creationdate>2023-04</creationdate><rights>http://arxiv.org/licenses/nonexclusive-distrib/1.0</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,776,881</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/2304.09542$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2304.09542$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Sun, Weiwei</creatorcontrib><creatorcontrib>Yan, Lingyong</creatorcontrib><creatorcontrib>Ma, Xinyu</creatorcontrib><creatorcontrib>Wang, Shuaiqiang</creatorcontrib><creatorcontrib>Ren, Pengjie</creatorcontrib><creatorcontrib>Chen, Zhumin</creatorcontrib><creatorcontrib>Yin, Dawei</creatorcontrib><creatorcontrib>Ren, Zhaochun</creatorcontrib><title>Is ChatGPT Good at Search? Investigating Large Language Models as Re-Ranking Agents</title><description>Large Language Models (LLMs) have demonstrated remarkable zero-shot generalization across various language-related tasks, including search engines. However, existing work utilizes the generative ability of LLMs for Information Retrieval (IR) rather than direct passage ranking. The discrepancy between the pre-training objectives of LLMs and the ranking objective poses another challenge. In this paper, we first investigate generative LLMs such as ChatGPT and GPT-4 for relevance ranking in IR. Surprisingly, our experiments reveal that properly instructed LLMs can deliver competitive, even superior results to state-of-the-art supervised methods on popular IR benchmarks. Furthermore, to address concerns about data contamination of LLMs, we collect a new test set called NovelEval, based on the latest knowledge and aiming to verify the model's ability to rank unknown knowledge. Finally, to improve efficiency in real-world applications, we delve into the potential for distilling the ranking capabilities of ChatGPT into small specialized models using a permutation distillation scheme. Our evaluation results turn out that a distilled 440M model outperforms a 3B supervised model on the BEIR benchmark. The code to reproduce our results is available at www.github.com/sunnweiwei/RankGPT.</description><subject>Computer Science - Computation and Language</subject><subject>Computer Science - Information Retrieval</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2023</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNotj0FOwzAURL1hgQoHYIUvkGDHdpysUBVBiJQK1GYffTs_bkRxkB0quD1tYTMzi6eRHiF3nKWyUIo9QPiejmkmmExZqWR2TXZNpNUelvqto_U8DxQWukMIdv9IG3_EuEwOlsk72kJweErvvuA0NvOAh0gh0i0mW_DvZ2bt0C_xhlyNcIh4-98r0j0_ddVL0r7WTbVuE8h1lijLFCs0h1KJUaAWgjPLrTIMEWSuOFo9lkYPhcmVzezIB61QajMIaYzhYkXu_24vVv1nmD4g_PRnu_5iJ34BDx5JVA</recordid><startdate>20230419</startdate><enddate>20230419</enddate><creator>Sun, Weiwei</creator><creator>Yan, Lingyong</creator><creator>Ma, Xinyu</creator><creator>Wang, Shuaiqiang</creator><creator>Ren, Pengjie</creator><creator>Chen, Zhumin</creator><creator>Yin, Dawei</creator><creator>Ren, Zhaochun</creator><scope>AKY</scope><scope>GOX</scope></search><sort><creationdate>20230419</creationdate><title>Is ChatGPT Good at Search? Investigating Large Language Models as Re-Ranking Agents</title><author>Sun, Weiwei ; Yan, Lingyong ; Ma, Xinyu ; Wang, Shuaiqiang ; Ren, Pengjie ; Chen, Zhumin ; Yin, Dawei ; Ren, Zhaochun</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a672-5c050871a953f3e73310c1c5b0eea4651ec7f9b7d8b65c2cf1d75e47bd34bbb13</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2023</creationdate><topic>Computer Science - Computation and Language</topic><topic>Computer Science - Information Retrieval</topic><toplevel>online_resources</toplevel><creatorcontrib>Sun, Weiwei</creatorcontrib><creatorcontrib>Yan, Lingyong</creatorcontrib><creatorcontrib>Ma, Xinyu</creatorcontrib><creatorcontrib>Wang, Shuaiqiang</creatorcontrib><creatorcontrib>Ren, Pengjie</creatorcontrib><creatorcontrib>Chen, Zhumin</creatorcontrib><creatorcontrib>Yin, Dawei</creatorcontrib><creatorcontrib>Ren, Zhaochun</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Sun, Weiwei</au><au>Yan, Lingyong</au><au>Ma, Xinyu</au><au>Wang, Shuaiqiang</au><au>Ren, Pengjie</au><au>Chen, Zhumin</au><au>Yin, Dawei</au><au>Ren, Zhaochun</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Is ChatGPT Good at Search? Investigating Large Language Models as Re-Ranking Agents</atitle><date>2023-04-19</date><risdate>2023</risdate><abstract>Large Language Models (LLMs) have demonstrated remarkable zero-shot generalization across various language-related tasks, including search engines. However, existing work utilizes the generative ability of LLMs for Information Retrieval (IR) rather than direct passage ranking. The discrepancy between the pre-training objectives of LLMs and the ranking objective poses another challenge. In this paper, we first investigate generative LLMs such as ChatGPT and GPT-4 for relevance ranking in IR. Surprisingly, our experiments reveal that properly instructed LLMs can deliver competitive, even superior results to state-of-the-art supervised methods on popular IR benchmarks. Furthermore, to address concerns about data contamination of LLMs, we collect a new test set called NovelEval, based on the latest knowledge and aiming to verify the model's ability to rank unknown knowledge. Finally, to improve efficiency in real-world applications, we delve into the potential for distilling the ranking capabilities of ChatGPT into small specialized models using a permutation distillation scheme. Our evaluation results turn out that a distilled 440M model outperforms a 3B supervised model on the BEIR benchmark. The code to reproduce our results is available at www.github.com/sunnweiwei/RankGPT.</abstract><doi>10.48550/arxiv.2304.09542</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.2304.09542
ispartof
issn
language eng
recordid cdi_arxiv_primary_2304_09542
source arXiv.org
subjects Computer Science - Computation and Language
Computer Science - Information Retrieval
title Is ChatGPT Good at Search? Investigating Large Language Models as Re-Ranking Agents
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-02-08T07%3A11%3A34IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-arxiv_GOX&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=Is%20ChatGPT%20Good%20at%20Search?%20Investigating%20Large%20Language%20Models%20as%20Re-Ranking%20Agents&rft.au=Sun,%20Weiwei&rft.date=2023-04-19&rft_id=info:doi/10.48550/arxiv.2304.09542&rft_dat=%3Carxiv_GOX%3E2304_09542%3C/arxiv_GOX%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_id=info:pmid/&rfr_iscdi=true