@inproceedings{han-etal-2026-illuminating,
title = "Illuminating the Path: Reachability-Aware Supervision Curation for {LLM}-Based Recommendation",
author = "Han, Donghee and
Jeong, Jiwon and
Song, Hwanjun and
Yi, Mun Yong",
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.552/",
pages = "12665--12686",
ISBN = "979-8-89176-459-0",
abstract = "Eliciting high-quality reasoning trajectories to supervise Large Language Models (LLMs) in sequential recommendation is notoriously difficult. In massive item pools, standard data curation methods such as autonomous exploration or rejection sampling suffer from severe sparsity. LLMs' context-utilization bias exacerbate this challenge by overusing noisy interaction histories, yielding diffuse and weakly grounded reasoning paths that frequently fail to retrieve the target item. To overcome this data curation bottleneck, we propose GEARec (Grounded Reasoning Curation for Anchor-based Recommendation), a framework based on Dual-Perspective Anchor Extraction and Hybrid Reasoning Generation. By combining the semantic strengths of LLMs with collaborative filtering (CF) signals, GEARec identifies a high-relevance anchor item from the user{'}s history to construct highly focused, reachability-aware reasoning trajectories for supervised fine-tuning. Experiments on multiple datasets demonstrate that GEARec effectively curates superior supervision signals, reducing context overuse and consistently outperforming strong baselines in recommendation accuracy."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="han-etal-2026-illuminating">
<titleInfo>
<title>Illuminating the Path: Reachability-Aware Supervision Curation for LLM-Based Recommendation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Donghee</namePart>
<namePart type="family">Han</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiwon</namePart>
<namePart type="family">Jeong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hwanjun</namePart>
<namePart type="family">Song</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mun</namePart>
<namePart type="given">Yong</namePart>
<namePart type="family">Yi</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>Eliciting high-quality reasoning trajectories to supervise Large Language Models (LLMs) in sequential recommendation is notoriously difficult. In massive item pools, standard data curation methods such as autonomous exploration or rejection sampling suffer from severe sparsity. LLMs’ context-utilization bias exacerbate this challenge by overusing noisy interaction histories, yielding diffuse and weakly grounded reasoning paths that frequently fail to retrieve the target item. To overcome this data curation bottleneck, we propose GEARec (Grounded Reasoning Curation for Anchor-based Recommendation), a framework based on Dual-Perspective Anchor Extraction and Hybrid Reasoning Generation. By combining the semantic strengths of LLMs with collaborative filtering (CF) signals, GEARec identifies a high-relevance anchor item from the user’s history to construct highly focused, reachability-aware reasoning trajectories for supervised fine-tuning. Experiments on multiple datasets demonstrate that GEARec effectively curates superior supervision signals, reducing context overuse and consistently outperforming strong baselines in recommendation accuracy.</abstract>
<identifier type="citekey">han-etal-2026-illuminating</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.552/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>12665</start>
<end>12686</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Illuminating the Path: Reachability-Aware Supervision Curation for LLM-Based Recommendation
%A Han, Donghee
%A Jeong, Jiwon
%A Song, Hwanjun
%A Yi, Mun Yong
%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 han-etal-2026-illuminating
%X Eliciting high-quality reasoning trajectories to supervise Large Language Models (LLMs) in sequential recommendation is notoriously difficult. In massive item pools, standard data curation methods such as autonomous exploration or rejection sampling suffer from severe sparsity. LLMs’ context-utilization bias exacerbate this challenge by overusing noisy interaction histories, yielding diffuse and weakly grounded reasoning paths that frequently fail to retrieve the target item. To overcome this data curation bottleneck, we propose GEARec (Grounded Reasoning Curation for Anchor-based Recommendation), a framework based on Dual-Perspective Anchor Extraction and Hybrid Reasoning Generation. By combining the semantic strengths of LLMs with collaborative filtering (CF) signals, GEARec identifies a high-relevance anchor item from the user’s history to construct highly focused, reachability-aware reasoning trajectories for supervised fine-tuning. Experiments on multiple datasets demonstrate that GEARec effectively curates superior supervision signals, reducing context overuse and consistently outperforming strong baselines in recommendation accuracy.
%U https://aclanthology.org/2026.emnlp-main.552/
%P 12665-12686
Markdown (Informal)
[Illuminating the Path: Reachability-Aware Supervision Curation for LLM-Based Recommendation](https://aclanthology.org/2026.emnlp-main.552/) (Han et al., EMNLP 2026)
ACL