@inproceedings{pang-etal-2026-ica,
title = "{ICA}: Information-Aware Credit Assignment for Visually Grounded Long-Horizon Information-Seeking Agents",
author = "Pang, Cong and
YI, Yujie and
Feng, Xuyu and
Su, Jiaqi and
Chen, Zixuan and
Hong, Jiawei and
Yao, Tiankuo and
Yuan, Nang and
Luo, Jiapeng and
Lu, Lewei and
Lou, Xin",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.327/",
pages = "7541--7556",
ISBN = "979-8-89176-459-0",
abstract = "Long-horizon reinforcement learning for information seeking agents remains difficult because terminal rewards reveal whether the final answer is correct, but not which acquired information enabled it. This difficulty is amplified by text-derived webpage observations, where parsing, truncation, and summarization often produce incomplete and unstable content representations across trajectories. We propose an evidence-centric framework for web agent learning that represents information acquired through tools as identifiable units for comparison across trajectories. In particular, fetched webpages are represented as rendered snapshots, preserving layout and multimodal content as stable content-level observations. Building on these units, we introduce Information-Aware Credit Assignment, a post hoc reward propagation method that estimates turn-level utility scores from rollout success rates and assigns dense rewards to intermediate steps that introduced high-utility information. Integrated with GSPO, our method consistently improves performance on BrowseComp, GAIA, Xbench-DS, and Seal-0. Code and datasets will be released at [https://github.com/pc-inno/ICA{\_}MM{\_}deepsearch](https://github.com/pc-inno/ICA{\_}MM{\_}deepsearch)."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="pang-etal-2026-ica">
<titleInfo>
<title>ICA: Information-Aware Credit Assignment for Visually Grounded Long-Horizon Information-Seeking Agents</title>
</titleInfo>
<name type="personal">
<namePart type="given">Cong</namePart>
<namePart type="family">Pang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yujie</namePart>
<namePart type="family">YI</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xuyu</namePart>
<namePart type="family">Feng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiaqi</namePart>
<namePart type="family">Su</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zixuan</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiawei</namePart>
<namePart type="family">Hong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tiankuo</namePart>
<namePart type="family">Yao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nang</namePart>
<namePart type="family">Yuan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiapeng</namePart>
<namePart type="family">Luo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lewei</namePart>
<namePart type="family">Lu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xin</namePart>
<namePart type="family">Lou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Long-horizon reinforcement learning for information seeking agents remains difficult because terminal rewards reveal whether the final answer is correct, but not which acquired information enabled it. This difficulty is amplified by text-derived webpage observations, where parsing, truncation, and summarization often produce incomplete and unstable content representations across trajectories. We propose an evidence-centric framework for web agent learning that represents information acquired through tools as identifiable units for comparison across trajectories. In particular, fetched webpages are represented as rendered snapshots, preserving layout and multimodal content as stable content-level observations. Building on these units, we introduce Information-Aware Credit Assignment, a post hoc reward propagation method that estimates turn-level utility scores from rollout success rates and assigns dense rewards to intermediate steps that introduced high-utility information. Integrated with GSPO, our method consistently improves performance on BrowseComp, GAIA, Xbench-DS, and Seal-0. Code and datasets will be released at [https://github.com/pc-inno/ICA_MM_deepsearch](https://github.com/pc-inno/ICA_MM_deepsearch).</abstract>
<identifier type="citekey">pang-etal-2026-ica</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.327/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>7541</start>
<end>7556</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T ICA: Information-Aware Credit Assignment for Visually Grounded Long-Horizon Information-Seeking Agents
%A Pang, Cong
%A YI, Yujie
%A Feng, Xuyu
%A Su, Jiaqi
%A Chen, Zixuan
%A Hong, Jiawei
%A Yao, Tiankuo
%A Yuan, Nang
%A Luo, Jiapeng
%A Lu, Lewei
%A Lou, Xin
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F pang-etal-2026-ica
%X Long-horizon reinforcement learning for information seeking agents remains difficult because terminal rewards reveal whether the final answer is correct, but not which acquired information enabled it. This difficulty is amplified by text-derived webpage observations, where parsing, truncation, and summarization often produce incomplete and unstable content representations across trajectories. We propose an evidence-centric framework for web agent learning that represents information acquired through tools as identifiable units for comparison across trajectories. In particular, fetched webpages are represented as rendered snapshots, preserving layout and multimodal content as stable content-level observations. Building on these units, we introduce Information-Aware Credit Assignment, a post hoc reward propagation method that estimates turn-level utility scores from rollout success rates and assigns dense rewards to intermediate steps that introduced high-utility information. Integrated with GSPO, our method consistently improves performance on BrowseComp, GAIA, Xbench-DS, and Seal-0. Code and datasets will be released at [https://github.com/pc-inno/ICA_MM_deepsearch](https://github.com/pc-inno/ICA_MM_deepsearch).
%U https://aclanthology.org/2026.emnlp-main.327/
%P 7541-7556
Markdown (Informal)
[ICA: Information-Aware Credit Assignment for Visually Grounded Long-Horizon Information-Seeking Agents](https://aclanthology.org/2026.emnlp-main.327/) (Pang et al., EMNLP 2026)
ACL
- Cong Pang, Yujie YI, Xuyu Feng, Jiaqi Su, Zixuan Chen, Jiawei Hong, Tiankuo Yao, Nang Yuan, Jiapeng Luo, Lewei Lu, and Xin Lou. 2026. ICA: Information-Aware Credit Assignment for Visually Grounded Long-Horizon Information-Seeking Agents. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 7541–7556, Budapest, Hungary. Association for Computational Linguistics.