@inproceedings{boytsov-etal-2025-positional,
title = "Positional Bias in Long-Document Ranking: Impact, Assessment, and Mitigation",
author = "Boytsov, Leonid and
Akinpelu, David and
Katyal, Nipun and
Lin, Tianyi and
Gao, Fangwei and
Zhao, Yutian and
Huang, Jeffrey and
Nyberg, Eric",
editor = "Inui, Kentaro and
Sakti, Sakriani and
Wang, Haofen and
Wong, Derek F. and
Bhattacharyya, Pushpak and
Banerjee, Biplab and
Ekbal, Asif and
Chakraborty, Tanmoy and
Singh, Dhirendra Pratap",
booktitle = "Proceedings of the 14th International Joint Conference on Natural Language Processing and the 4th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics",
month = dec,
year = "2025",
address = "Mumbai, India",
publisher = "The Asian Federation of Natural Language Processing and The Association for Computational Linguistics",
url = "https://aclanthology.org/2025.ijcnlp-long.46/",
pages = "824--856",
ISBN = "979-8-89176-298-5",
abstract = "We tested over 20 Transformer models for ranking long documents (including recent LongP models trained with FlashAttention and RankGPT models ``powered'' by OpenAI and Anthropic cloud APIs).We compared them with the simple FirstP baseline, which applied the same model to truncated input (up to 512 tokens).On MS MARCO, TREC DL, and Robust04 no long-document model outperformed FirstP by more than 5{\%} (on average).We hypothesized that this lack of improvement is not due to inherent model limitations,but due to benchmark positional bias (most relevant passages tend to occur early in documents),which is known to exist in MS MARCO.To confirm this, we analyzed positional relevance distributions across four long-document corpora (with six query sets) and observed the same early-position bias.Surprisingly, we also found bias in six BEIR collections, which are typically categorized asshort-document datasets.We then introduced a new diagnostic dataset, MS MARCO FarRelevant, where relevant spans were deliberately placed beyond the first 512 tokens.On this dataset, many long-context models{---}including RankGPT{---}performed at random-baseline level, suggesting overfitting to positional bias.We also experimented with debiasing training data, but with limited success.Our findings (1) highlight the need for careful benchmark design in evaluating long-context models for document ranking, (2) identify model types that are more robust to positional bias, and (3) motivate further work on approaches to debias training data.We release our code and data to support further research."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="boytsov-etal-2025-positional">
<titleInfo>
<title>Positional Bias in Long-Document Ranking: Impact, Assessment, and Mitigation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Leonid</namePart>
<namePart type="family">Boytsov</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">David</namePart>
<namePart type="family">Akinpelu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nipun</namePart>
<namePart type="family">Katyal</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tianyi</namePart>
<namePart type="family">Lin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Fangwei</namePart>
<namePart type="family">Gao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yutian</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jeffrey</namePart>
<namePart type="family">Huang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Eric</namePart>
<namePart type="family">Nyberg</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-12</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 14th International Joint Conference on Natural Language Processing and the 4th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics</title>
</titleInfo>
<name type="personal">
<namePart type="given">Kentaro</namePart>
<namePart type="family">Inui</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sakriani</namePart>
<namePart type="family">Sakti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Haofen</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Derek</namePart>
<namePart type="given">F</namePart>
<namePart type="family">Wong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Pushpak</namePart>
<namePart type="family">Bhattacharyya</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Biplab</namePart>
<namePart type="family">Banerjee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Asif</namePart>
<namePart type="family">Ekbal</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tanmoy</namePart>
<namePart type="family">Chakraborty</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dhirendra</namePart>
<namePart type="given">Pratap</namePart>
<namePart type="family">Singh</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>The Asian Federation of Natural Language Processing and The Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Mumbai, India</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-298-5</identifier>
</relatedItem>
<abstract>We tested over 20 Transformer models for ranking long documents (including recent LongP models trained with FlashAttention and RankGPT models “powered” by OpenAI and Anthropic cloud APIs).We compared them with the simple FirstP baseline, which applied the same model to truncated input (up to 512 tokens).On MS MARCO, TREC DL, and Robust04 no long-document model outperformed FirstP by more than 5% (on average).We hypothesized that this lack of improvement is not due to inherent model limitations,but due to benchmark positional bias (most relevant passages tend to occur early in documents),which is known to exist in MS MARCO.To confirm this, we analyzed positional relevance distributions across four long-document corpora (with six query sets) and observed the same early-position bias.Surprisingly, we also found bias in six BEIR collections, which are typically categorized asshort-document datasets.We then introduced a new diagnostic dataset, MS MARCO FarRelevant, where relevant spans were deliberately placed beyond the first 512 tokens.On this dataset, many long-context models—including RankGPT—performed at random-baseline level, suggesting overfitting to positional bias.We also experimented with debiasing training data, but with limited success.Our findings (1) highlight the need for careful benchmark design in evaluating long-context models for document ranking, (2) identify model types that are more robust to positional bias, and (3) motivate further work on approaches to debias training data.We release our code and data to support further research.</abstract>
<identifier type="citekey">boytsov-etal-2025-positional</identifier>
<location>
<url>https://aclanthology.org/2025.ijcnlp-long.46/</url>
</location>
<part>
<date>2025-12</date>
<extent unit="page">
<start>824</start>
<end>856</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Positional Bias in Long-Document Ranking: Impact, Assessment, and Mitigation
%A Boytsov, Leonid
%A Akinpelu, David
%A Katyal, Nipun
%A Lin, Tianyi
%A Gao, Fangwei
%A Zhao, Yutian
%A Huang, Jeffrey
%A Nyberg, Eric
%Y Inui, Kentaro
%Y Sakti, Sakriani
%Y Wang, Haofen
%Y Wong, Derek F.
%Y Bhattacharyya, Pushpak
%Y Banerjee, Biplab
%Y Ekbal, Asif
%Y Chakraborty, Tanmoy
%Y Singh, Dhirendra Pratap
%S Proceedings of the 14th International Joint Conference on Natural Language Processing and the 4th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics
%D 2025
%8 December
%I The Asian Federation of Natural Language Processing and The Association for Computational Linguistics
%C Mumbai, India
%@ 979-8-89176-298-5
%F boytsov-etal-2025-positional
%X We tested over 20 Transformer models for ranking long documents (including recent LongP models trained with FlashAttention and RankGPT models “powered” by OpenAI and Anthropic cloud APIs).We compared them with the simple FirstP baseline, which applied the same model to truncated input (up to 512 tokens).On MS MARCO, TREC DL, and Robust04 no long-document model outperformed FirstP by more than 5% (on average).We hypothesized that this lack of improvement is not due to inherent model limitations,but due to benchmark positional bias (most relevant passages tend to occur early in documents),which is known to exist in MS MARCO.To confirm this, we analyzed positional relevance distributions across four long-document corpora (with six query sets) and observed the same early-position bias.Surprisingly, we also found bias in six BEIR collections, which are typically categorized asshort-document datasets.We then introduced a new diagnostic dataset, MS MARCO FarRelevant, where relevant spans were deliberately placed beyond the first 512 tokens.On this dataset, many long-context models—including RankGPT—performed at random-baseline level, suggesting overfitting to positional bias.We also experimented with debiasing training data, but with limited success.Our findings (1) highlight the need for careful benchmark design in evaluating long-context models for document ranking, (2) identify model types that are more robust to positional bias, and (3) motivate further work on approaches to debias training data.We release our code and data to support further research.
%U https://aclanthology.org/2025.ijcnlp-long.46/
%P 824-856
Markdown (Informal)
[Positional Bias in Long-Document Ranking: Impact, Assessment, and Mitigation](https://aclanthology.org/2025.ijcnlp-long.46/) (Boytsov et al., IJCNLP-AACL 2025)
ACL
- Leonid Boytsov, David Akinpelu, Nipun Katyal, Tianyi Lin, Fangwei Gao, Yutian Zhao, Jeffrey Huang, and Eric Nyberg. 2025. Positional Bias in Long-Document Ranking: Impact, Assessment, and Mitigation. In Proceedings of the 14th International Joint Conference on Natural Language Processing and the 4th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics, pages 824–856, Mumbai, India. The Asian Federation of Natural Language Processing and The Association for Computational Linguistics.