@inproceedings{gan-etal-2026-kg,
title = "{KG}-{RLVR}: Reinforcing {LLM} Reasoning with Knowledge Graph-Derived Process Rewards",
author = "Gan, Aoran and
Zhang, Guoxuan and
Li, Ya and
Wang, Qiangbing and
Liu, Quan and
Liu, Qi and
Hu, Guoping",
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.890/",
pages = "20538--20554",
ISBN = "979-8-89176-459-0",
abstract = "As large language models (LLMs) are increasingly adapted from general-purpose systems into domain-specific problem solvers, a central challenge is how to provide reliable process-level guidance for knowledge-intensive reasoning. High-quality reasoning supervision, such as annotated reasoning trajectories or externally generated rationales, is costly to obtain, difficult to verify, and hard to scale across domains. In this paper, we propose KG-RLVR, a lightweight framework that reframes KG-enhanced LLM post-training as verifiable reward construction from traceable structured evidence. KG-RLVR retrieves evidence from external knowledge graphs, consolidates noisy graph neighborhoods into representative evidence units, and converts these units into locally computable process rewards. The reward encourages the model to cover directly relevant evidence while selectively integrating less direct but potentially useful auxiliary evidence, thereby guiding reasoning traces without requiring externally authored process demonstrations. Experiments on knowledge-intensive medical and multi-domain QA benchmarks show that KG-RLVR consistently improves over vanilla RL, demonstrating that KG-derived process rewards provide practical and scalable supervision for reasoning-oriented LLM post-training."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="gan-etal-2026-kg">
<titleInfo>
<title>KG-RLVR: Reinforcing LLM Reasoning with Knowledge Graph-Derived Process Rewards</title>
</titleInfo>
<name type="personal">
<namePart type="given">Aoran</namePart>
<namePart type="family">Gan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Guoxuan</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ya</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Qiangbing</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Quan</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Qi</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Guoping</namePart>
<namePart type="family">Hu</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>As large language models (LLMs) are increasingly adapted from general-purpose systems into domain-specific problem solvers, a central challenge is how to provide reliable process-level guidance for knowledge-intensive reasoning. High-quality reasoning supervision, such as annotated reasoning trajectories or externally generated rationales, is costly to obtain, difficult to verify, and hard to scale across domains. In this paper, we propose KG-RLVR, a lightweight framework that reframes KG-enhanced LLM post-training as verifiable reward construction from traceable structured evidence. KG-RLVR retrieves evidence from external knowledge graphs, consolidates noisy graph neighborhoods into representative evidence units, and converts these units into locally computable process rewards. The reward encourages the model to cover directly relevant evidence while selectively integrating less direct but potentially useful auxiliary evidence, thereby guiding reasoning traces without requiring externally authored process demonstrations. Experiments on knowledge-intensive medical and multi-domain QA benchmarks show that KG-RLVR consistently improves over vanilla RL, demonstrating that KG-derived process rewards provide practical and scalable supervision for reasoning-oriented LLM post-training.</abstract>
<identifier type="citekey">gan-etal-2026-kg</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.890/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>20538</start>
<end>20554</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T KG-RLVR: Reinforcing LLM Reasoning with Knowledge Graph-Derived Process Rewards
%A Gan, Aoran
%A Zhang, Guoxuan
%A Li, Ya
%A Wang, Qiangbing
%A Liu, Quan
%A Liu, Qi
%A Hu, Guoping
%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 gan-etal-2026-kg
%X As large language models (LLMs) are increasingly adapted from general-purpose systems into domain-specific problem solvers, a central challenge is how to provide reliable process-level guidance for knowledge-intensive reasoning. High-quality reasoning supervision, such as annotated reasoning trajectories or externally generated rationales, is costly to obtain, difficult to verify, and hard to scale across domains. In this paper, we propose KG-RLVR, a lightweight framework that reframes KG-enhanced LLM post-training as verifiable reward construction from traceable structured evidence. KG-RLVR retrieves evidence from external knowledge graphs, consolidates noisy graph neighborhoods into representative evidence units, and converts these units into locally computable process rewards. The reward encourages the model to cover directly relevant evidence while selectively integrating less direct but potentially useful auxiliary evidence, thereby guiding reasoning traces without requiring externally authored process demonstrations. Experiments on knowledge-intensive medical and multi-domain QA benchmarks show that KG-RLVR consistently improves over vanilla RL, demonstrating that KG-derived process rewards provide practical and scalable supervision for reasoning-oriented LLM post-training.
%U https://aclanthology.org/2026.emnlp-main.890/
%P 20538-20554
Markdown (Informal)
[KG-RLVR: Reinforcing LLM Reasoning with Knowledge Graph-Derived Process Rewards](https://aclanthology.org/2026.emnlp-main.890/) (Gan et al., EMNLP 2026)
ACL