@inproceedings{wang-etal-2025-gradient,
title = "Gradient-guided Attention Map Editing: Towards Efficient Contextual Hallucination Mitigation",
author = "Wang, Yu and
Zhang, Jiaxin and
Gao, Xiang and
Cui, Wendi and
Li, Peng and
Das, Kamalika",
editor = "Chiruzzo, Luis and
Ritter, Alan and
Wang, Lu",
booktitle = "Findings of the Association for Computational Linguistics: NAACL 2025",
month = apr,
year = "2025",
address = "Albuquerque, New Mexico",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.findings-naacl.458/",
doi = "10.18653/v1/2025.findings-naacl.458",
pages = "8206--8217",
ISBN = "979-8-89176-195-7",
abstract = "In tasks such as summarization and open-book question answering (QA), Large Language Models (LLMs) frequently experience ``contextual hallucination'', where they generate irrelevant or incorrect responses despite having access to accurate information in the input. This issue often stems from the models' propensity to prioritize self-generated content over input context, leading to a disregard for pertinent details. To address this challenge, we introduce, Guided Attention Map Editing (GAME), an innovative approach that dynamically adjusts attention maps to enhance contextual relevance. During inference, GAME employs a trained classifier to identify attention maps likely to induce hallucinations and implements targeted interventions. These interventions, guided by gradient-informed ``edit directions'', strategically redistribute attention weights across various heads to efficiently mitigate hallucination. Extensive evaluations on challenging summarization and open-book QA tasks demonstrate that GAME consistently and significantly reduces hallucinations across diverse open-source models, thereby improving the reliability and applicability of LLMs."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="wang-etal-2025-gradient">
<titleInfo>
<title>Gradient-guided Attention Map Editing: Towards Efficient Contextual Hallucination Mitigation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yu</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiaxin</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiang</namePart>
<namePart type="family">Gao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wendi</namePart>
<namePart type="family">Cui</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Peng</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kamalika</namePart>
<namePart type="family">Das</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-04</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: NAACL 2025</title>
</titleInfo>
<name type="personal">
<namePart type="given">Luis</namePart>
<namePart type="family">Chiruzzo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alan</namePart>
<namePart type="family">Ritter</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lu</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Albuquerque, New Mexico</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-195-7</identifier>
</relatedItem>
<abstract>In tasks such as summarization and open-book question answering (QA), Large Language Models (LLMs) frequently experience “contextual hallucination”, where they generate irrelevant or incorrect responses despite having access to accurate information in the input. This issue often stems from the models’ propensity to prioritize self-generated content over input context, leading to a disregard for pertinent details. To address this challenge, we introduce, Guided Attention Map Editing (GAME), an innovative approach that dynamically adjusts attention maps to enhance contextual relevance. During inference, GAME employs a trained classifier to identify attention maps likely to induce hallucinations and implements targeted interventions. These interventions, guided by gradient-informed “edit directions”, strategically redistribute attention weights across various heads to efficiently mitigate hallucination. Extensive evaluations on challenging summarization and open-book QA tasks demonstrate that GAME consistently and significantly reduces hallucinations across diverse open-source models, thereby improving the reliability and applicability of LLMs.</abstract>
<identifier type="citekey">wang-etal-2025-gradient</identifier>
<identifier type="doi">10.18653/v1/2025.findings-naacl.458</identifier>
<location>
<url>https://aclanthology.org/2025.findings-naacl.458/</url>
</location>
<part>
<date>2025-04</date>
<extent unit="page">
<start>8206</start>
<end>8217</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Gradient-guided Attention Map Editing: Towards Efficient Contextual Hallucination Mitigation
%A Wang, Yu
%A Zhang, Jiaxin
%A Gao, Xiang
%A Cui, Wendi
%A Li, Peng
%A Das, Kamalika
%Y Chiruzzo, Luis
%Y Ritter, Alan
%Y Wang, Lu
%S Findings of the Association for Computational Linguistics: NAACL 2025
%D 2025
%8 April
%I Association for Computational Linguistics
%C Albuquerque, New Mexico
%@ 979-8-89176-195-7
%F wang-etal-2025-gradient
%X In tasks such as summarization and open-book question answering (QA), Large Language Models (LLMs) frequently experience “contextual hallucination”, where they generate irrelevant or incorrect responses despite having access to accurate information in the input. This issue often stems from the models’ propensity to prioritize self-generated content over input context, leading to a disregard for pertinent details. To address this challenge, we introduce, Guided Attention Map Editing (GAME), an innovative approach that dynamically adjusts attention maps to enhance contextual relevance. During inference, GAME employs a trained classifier to identify attention maps likely to induce hallucinations and implements targeted interventions. These interventions, guided by gradient-informed “edit directions”, strategically redistribute attention weights across various heads to efficiently mitigate hallucination. Extensive evaluations on challenging summarization and open-book QA tasks demonstrate that GAME consistently and significantly reduces hallucinations across diverse open-source models, thereby improving the reliability and applicability of LLMs.
%R 10.18653/v1/2025.findings-naacl.458
%U https://aclanthology.org/2025.findings-naacl.458/
%U https://doi.org/10.18653/v1/2025.findings-naacl.458
%P 8206-8217
Markdown (Informal)
[Gradient-guided Attention Map Editing: Towards Efficient Contextual Hallucination Mitigation](https://aclanthology.org/2025.findings-naacl.458/) (Wang et al., Findings 2025)
ACL