ICLR 2026 | LLM×Graph论文总结【LLM4Graph与Graph4LLM】

ICLR 2026将在2026年4月23日到27日于巴西里约热内卢（Rio de Janeiro, Brazil）举行。ICLR 2025共有19,000多篇投稿，录用5,359篇，录取率28.18%。

本文总结了2026 ICLR上有关LLM Graph的相关论文。总计29篇，如有疏漏，欢迎补充。

观察：LLM Graph统计值

其中均分≥6的有4篇，其中。

笔者将LLM和Graph结合的工作分为两大类，一类是LLM4Graph，即LLM做图任务。其中细分了：文本属性图（Text-Attributed Graph, TAG）,知识图谱（KG），图基础模型（GFM）,AI4Science，图上的推理和理解任务。另外一类是利用Graph4LLM，即利用图结构来增强LLM的能力。

LLM4Graph & TAG

1 Global-Recent Semantic Reasoning on Dynamic Text-Attributed Graphs with Large Language Models

链接：https://openreview.net/forum?id=puocvrFZRl

关键词：Dynamic text-attributed graph, graph learning, large language model

作者：Yunan Wang, Jianxin Li, Ziwei Zhang

分数：6, 2, 8, 6

信心：4, 5, 4, 4

均分：5.5

TL; DR：This paper propose DyGRASP, which captures the recent-global semantics inherent in dynamic text-attribute graphs with large language models.

2 GDGB: A Benchmark for Generative Dynamic Text-Attributed Graph Learning

链接：https://openreview.net/forum?id=5UFUHUC5qP

关键词：Dynamic Text-Attributed Graph, Dynamic Graph Generation

作者：Jie Peng, Jiarui Ji, Runlin Lei, Zhewei Wei, Yongchao Liu, Chuntao Hong

分数：4, 8, 4

信心：5, 3, 3

均分：5.333333333333333

TL; DR：We propose Generative DyTAGs Benchmark (GDGB), addressing poor textual quality in existing datasets and defining novel tasks (TDGG/IDGG) to advance robust and reproducible DyTAG generation research.

3 Robustness in Text-Attributed Graph Learning: Insights, Trade-offs, and New Defenses

链接：https://openreview.net/forum?id=CEJl0gN2gj

关键词：Graph Robustness, Graph Adversarial Attack, Text Attributed Graph, Large Language Model

作者：Runlin Lei, Lu Yi, Mingguo He, Pengyu Qiu, Zhewei Wei, Yongchao Liu, Chuntao Hong

分数：6, 4, 2

信心：4, 4, 4

均分：4.0

TL; DR：We propose a comprehensive evaluation of the robustness of predictors on text-attributed graphs.

LLM4Graph & KG

4 DAMR: Efficient and Adaptive Context-Aware Knowledge Graph Question Answering with LLM-Guided MCTS

链接：https://openreview.net/forum?id=mUx7WLC8q6

关键词：Knowledge Graphs, Question Answering, LLMs

作者：Yingxu Wang, Shiqi Fan, Mengzhu Wang, Siyang Gao, Chao Wang, Nan Yin

分数：4, 8, 6

信心：4, 4, 3

均分：6.0

5 Explore-on-Graph: Incentivizing Autonomous Exploration of Large Language Models on Knowledge Graphs with Path-refined Reward Modeling

链接：https://openreview.net/forum?id=NfuBj8jleE

关键词：Knowledge Graph, Large Language Models, Knowledge-enhanced reasoning, reinforcement learning

作者：Shiqi Yan, Yubo Chen, Ruiqi Zhou, Zhengxi Yao, Shuai Chen, Tianyi Zhang, Shijie Zhang, Wei-Qiang Zhang, Yongfeng Huang, Haixin Duan, Yunqi Zhang

分数：6, 8, 4, 4

信心：3, 4, 4, 3

均分：5.4

6 Plan-Answer-Refine-on-Graph: Structured Planning and Self-Refinement for Large Language Model Reasoning on Knowledge Graphs

链接：https://openreview.net/forum?id=g6XnP7Sgui

关键词：Knowledge Graphs, Large Language Models, Question Answering

作者：Yuxin Shi, Han Fu, Xiaoxue Ren, Chenghao Liu, Zhuo Li, Jianling Sun

分数：6, 4, 4, 6

信心：3, 5, 4, 5

均分：5.0

7 Knowledge Reasoning Language Model: Unifying Knowledge and Language for Inductive Knowledge Graph Reasoning

链接：https://openreview.net/forum?id=2g8EmFwNTB

关键词：Inductive Knowledge Graph Reasoning, Large Language Model, Knowledge Graph Foundation Model

作者：Xingrui Zhuo, Jiapu Wang, Gongqing Wu, Zhongyuan Wang, Jichen Zhang, Shirui Pan, Xindong Wu

分数：6, 4, 6, 4

信心：3, 4, 4, 3

均分：5.0

TL; DR：We propose an inductive knowledge graph reasoning foundation model that unifies structural knowledge and LLM, with significant zero-shot learning ability on unknown KGs

8 HGNet: Scalable Foundation Model for Automated Knowledge Graph Generation from Scientific Literature

链接：https://openreview.net/forum?id=NWd53rltx8

关键词：Knowledge Graphs, Representation Learning, Graph Neural Networks, Geometric Deep Learning, Scientific Text Mining

作者：Devvrat Joshi, Islem Rekik

分数：2, 8, 4

信心：3, 3, 4

均分：4.67

TL; DR：The first lightweight (<1B parameters), hierarchy-aware framework for building high-quality knowledge graphs from scientific research papers, significantly outperforming all state-of-the-art models in entity and relation extraction.

9 Flock: A Knowledge Graph Foundation Model via Learning on Random Walks

链接：https://openreview.net/forum?id=1cGOCIOKQd

关键词：knowledge graphs, link prediction, knowledge graph foundation models, invariance, equivariance, random walks

作者：Jinwoo Kim, Xingyue Huang, Krzysztof Olejniczak, Kyungbin Min, Michael M. Bronstein, Seunghoon Hong, Ismail Ilkan Ceylan

分数：8, 4, 6, 4

信心：3, 3, 4, 4

均分：5.5

TL; DR：We present Flock, a knowledge graph foundation model (KGFM) that uses random walks to achieve probabilistic node-relation equivariance and overcome the limitations of previous KGFMs.

LLM4Graph & GFM

10 [oral] Multi-Domain Transferable Graph Gluing for Building Graph Foundation Models

录用类型：oral

链接：https://openreview.net/forum?id=G3uNHQpP7J

关键词：Multi-domain graph pre-training, graph neural network, graph foundation model, Riemannian geometry

作者：Li Sun, Zhenhao Huang, Silei Chen, Lanxu Yang, Junda Ye, Sen Su, Philip S. Yu

分数：4, 8, 6, 6

信心：2, 4, 4, 4

均分：6.0

TL; DR：From differential geometry perspective, we present a novel framework that merges multi-domain graphs into a unified, smooth manifold with geometric consistency, enabling quantifiable transferability and geometric scaling behavior.

11 Bridging Input Feature Spaces Towards Graph Foundation Models

链接：https://openreview.net/forum?id=Dt4XAIKYbf

关键词：Graph Neural Networks, Graph Foundatin Models

作者：Moshe Eliasof, Krishna Sri Ipsit Mantri, Beatrice Bevilacqua, Bruno Ribeiro, Carola-Bibiane Sch枚nlieb

分数：10, 4, 2, 6

信心：4, 4, 4, 4

均分：5.5

TL; DR：We address the lack of a shared input space in graphs. We propose ALL-IN: map node features to a shared random space and build covariance-based representations invariant to feature permutations and orthogonal transforms, enabling zero-shot transfer.

12 G-reasoner: Foundation Models for Unified Reasoning over Graph-structured Knowledge

链接：https://openreview.net/forum?id=zJm9nmoahk

关键词：GraphRAG, RAG, LLM

作者：Linhao Luo, Zicheng Zhao, Junnan Liu, Zhangchi Qiu, Junnan Dong, Serge Panev, Chen Gong, Thuy-Trang Vu, Gholamreza Haffari, Dinh Phung, Alan Liew, Shirui Pan

分数：4, 6, 4, 8

信心：4, 3, 5, 3

均分：5.4

TL; DR：we present G-reasoner, a unified framework that integrates graph and language foundation models for reasoning over diverse graph-structured knowledge.

13 GNN-as-Judge: Unleashing the Power of LLMs for Graph Few-shot Semi-supervised Learning with GNN Feedback

链接：https://openreview.net/forum?id=nOlhDjNXKa

关键词：Large Language Models, Graph Neural Networks, Graph Few-shot Semi-supervised Learning

作者：Ruiyao Xu, Kaize Ding

分数：2, 4, 6, 8

信心：5, 4, 5, 3

均分：5.0

TL; DR：We propose GNN-as-Judge, a framework that leverages GNNs' feedback to select reliable pseudo-labels and a weakly supervised fine-tuning approach for tuning LLMs.

LLM4Graph & AI4Science

14 A Brain Graph Foundation Model: Pre-Training and Prompt-Tuning across Broad Atlases and Disorders

链接：https://openreview.net/forum?id=PeGHkAaRxs

关键词：Brain Graph Foundation Model, Functional Magnetic Resonance Imaging (fMRI), Neuroscience, Graph Pre-Training, Fine-Tuning, Prompt Learning

作者：Xinxu Wei, kanhao zhao, Yong Jiao, Lifang He, Yu Zhang

分数：6, 8, 4, 4

信心：4, 3, 4, 5

均分：5.5

15 GALAX: Graph-Augmented Language Model for Explainable Reinforcement-Guided Subgraph Reasoning in Precision Medicine

链接：https://openreview.net/forum?id=ADFXCeYXvR

关键词：Reinforcement Learning, Large Language Model (LLM), Text-Numeric Graph (TNG), Multi-Omics Integration, Explainability

作者：Heming Zhang, Di Huang, Wenyu Li, Michael Province, Yixin Chen, Philip Payne, Fuhai Li

分数：4, 6, 8, 2

信心：3, 2, 4, 4

均分：5.0

16 Entropy-Guided Dynamic Tokens for Graph-LLM Alignment in Molecular Understanding

链接：https://openreview.net/forum?id=yzwSzhqLpH

关键词：Multimodal Modeling, Graph–LLM Alignment, Molecule Understanding, Backbone-Free Tuning

作者：Zihao Jing, Qiuhao Zeng, Ruiyi Fang, Yan Sun, Boyu Wang, Pingzhao Hu

分数：2, 4, 6, 4

信心：4, 4, 5, 3

均分：4.0

TL; DR：EDT-Former: entropy-guided dynamic query tokens map molecular graphs to LLMs, capturing local and global structure features for comprehensive understanding and reasoning with backbone-free, connector-only training.

LLM4Graph & Graph Inference

17 GraphOmni: A Comprehensive and Extensible Benchmark Framework for Large Language Models on Graph-theoretic Tasks

链接：https://openreview.net/forum?id=dc8Kf2g0KC

关键词：LLM, Benchmark and Evaluation, Prompt Optimization

作者：Hao Xu, Xiangru Jian, Xinjian Zhao, Wei Pang, Chao Zhang, Suyuchen Wang, Qixin ZHANG, Zhengyuan Dong, Joao Monteiro, Bang Liu, Qiuzhuang Sun, Tianshu Yu

分数：6, 6, 6, 4

信心：3, 2, 5, 5

均分：5.4

18 [Oral] Actions Speak Louder than Prompts: A Large-Scale Study of LLMs for Graph Inference

录用类型：oral

链接：https://openreview.net/forum?id=MgJUj9Sk3C

关键词：Large Language Models, Prompting, In-Context Learning, Tool-augmented Reasoning, Text-rich Graphs

作者：Ben Finkelshtein, Silviu Cucerzan, Sujay Kumar Jauhar, Ryen White

分数：2, 6, 8, 6

信心：3, 4, 3, 2

均分：5.4

TL; DR：A comprehensive study of LLMs for node classification, providing a principled understanding of their capabilities in processing graph information that practitioners can apply in real-world tasks

19 <SO>: One LLM Token for Explicit Graph Structural Understanding

链接：https://openreview.net/forum?id=eXidGkRUFt

关键词：LLM for Graph, Graph Structure Learning, Structure Hallucination

作者：Jingyao Wu, Bin Lu, Zijun Di, Xiaoying Gan, Meng Jin, Luoyi Fu, Xinbing Wang, Chenghu Zhou

分数：4, 4, 4, 6

信心：4, 3, 3, 3

均分：4.5

Other GFM（AI4Science, KG）

20 HYPER: A Foundation Model for Inductive Link Prediction with Knowledge Hypergraphs

链接：https://openreview.net/forum?id=YLTQbMoAaX

关键词：Knowledge Hypergraph, Link Prediction, Graph Neural Networks, Foundation Models

作者：Xingyue Huang, Mikhail Galkin, Michael Bronstein, Ismail I Ceylan

分数：8, 6, 4, 2

信心：4, 3, 4, 4

均分：5.0

TL; DR：We develop the first foundation model over inductive link prediction with knowledge hypergraphs.

21 HEIST: A Graph Foundation Model for Spatial Transcriptomics and Proteomics Data

链接：https://openreview.net/forum?id=lK82jpa8jr

关键词：Pretrained models, Spatial transcriptomics, AI for Science

作者：Hiren Madhu, João Felipe Rocha, Tinglin Huang, Siddharth Viswanath, Smita Krishnaswamy, Rex Ying

分数：6, 4, 6, 4

信心：2, 4, 5, 3

均分：5.0

22 FragFM: Hierarchical Framework for Efficient Molecule Generation via Fragment-Level Discrete Flow Matching

链接：https://openreview.net/forum?id=tr6vRn2aPg

关键词：Molecular Graph Generation, Discrete Flow Matching, Fragment-Based Drug Discovery, Natural Product

作者：Joongwon Lee, Seonghwan Kim, Seokhyun Moon, Hyunwoo Kim, Woo Youn Kim

分数：6, 6, 4, 4

信心：3, 4, 3, 4

均分：5.0

TL; DR：We introduce FragFM, a novel hierarchical framework employing fragment‐level discrete flow matching for efficient molecular graph generation, along with a new molecular generative benchmark focused on natural products.

Graph4LLM

23 AtlasKV: Augmenting LLMs with Billion-Scale Knowledge Graphs in 20GB VRAM

链接：https://openreview.net/forum?id=6i1jVAYbHs

关键词：Large language model; Knowledge augmentation; Knowledge graph;

作者：Haoyu Huang, Hong Ting Tsang, Jiaxin Bai, Xi Peng, Gong Zhang, Yangqiu Song

分数：8, 8, 4, 4

信心：3, 4, 3, 4

均分：6.0

TL; DR：This paper proposes AtlasKV, a scalable, effective, and general way to augment LLMs with billion-scale KGs in less than 20GB GPU VRAM, where KG2KV and HiKVP are introduced to integrate KG triples at scale with sub-linear time and memory complexity.

24 VoG: Enhancing LLM Reasoning through Stepwise Verification on Knowledge Graphs

链接：https://openreview.net/forum?id=0RdAmwfVku

关键词：LLM reasoning, Knowledge Graphs, KG-enhanced LLM

作者：Wenxin Zhao, Jiachuan Wang, Yongqi Zhang, Shuangyin Li, Cheng Deng, Jun Wang, Lei Chen

分数：4, 4, 8, 6

信心：5, 5, 5, 4

均分：5.4

25 DAG-Math: Graph-Guided Mathematical Reasoning in LLMs

链接：https://openreview.net/forum?id=ylr6WArKQN

关键词：LLMs, mathematical reasoning, directed acyclic graphs

作者：Yuanhe Zhang, Ilja Kuzborskij, Jason Lee, Chenlei Leng, Fanghui Liu

分数：4, 6, 8, 6

信心：4, 3, 3, 2

均分：6.0

TL; DR：We propose a new pipeline by modeling CoT on directed acyclic graphs (DAGs), introduce the concept of logic closeness, and then precisely evaluates the mathematical reasoning ability of LLMs via the proposed DAG-MATH format.

26 GraphPlanner: Graph-Based Agentic Routing for LLMs

链接：https://openreview.net/forum?id=ZdGB7MNQDT

关键词：Agentic LLM, Memory utilization, Heterogeneous agents

作者：Tao Feng, Haozhen Zhang, Peixuan Han, Zijie Lei, Jiaxuan You

分数：4, 4, 8

信心：4, 3, 3

均分：5.33

TL; DR：GraphPlanner is a graph-based framework that enables agentic LLM routing by modeling cooperation and memory with reinforcement learning, achieving scalable, efficient, and generalizable routing.

27 Graph-of-Agents: A Graph-based Framework for Multi-Agent LLM Collaboration

链接：https://openreview.net/forum?id=34cANdsHKV

关键词：LLM Collaboration, Multi-Agent LLM

作者：Sukwon Yun, Jie Peng, Pingzhi Li, Wendong Fan, Jie Chen, James Y Zou, Guohao Li, Tianlong Chen

分数：4, 4, 6, 4

信心：2, 5, 3, 4

均分：4.5

28 GTool: Graph Enhanced Tool Planning with Large Language Model

链接：https://openreview.net/forum?id=bn47cqGQ7l

关键词：Tool Learning, Large Language Model, Graph Data Mining

作者：Wenjie Chen, Di Yao, Wenbin Li, Xuying Meng, Chang Gong, Jingping Bi

分数：4, 6, 6, 4

信心：4, 4, 4, 4

均分：5.0

TL; DR：We propose GTool, which is the first work aiming to enhance the tool planning ability of LLMs under incomplete tool dependencies.

29 CoMem: Compositional Concept-Graph Memory for Continual Vision–Language Learning

链接：https://openreview.net/forum?id=xp7wDU9JBW

关键词：VLM, Vision Language Learning, Continual Learning

作者：Heng Zhou, Jing Tang, Juheng zhang, Yanshu Li, Canran Xiao, Liwei Hou, Zong Ke, Jiawei Yao

分数：4, 6, 4, 6

信心：3, 2, 5, 3

均分：5.0

推荐阅读

ICLR 2026 | Rebuttal前 图基础模型(GFM)&文本属性图(TAG)高分论文

AAAI 2026 | 图基础模型(GFM)&文本属性图(TAG)论文总结

此公众号的文章皆系本人原创，辛苦码字不易！如需转载，引用请注明出处。如商用联系作者。

如果觉得有帮助还请分享，在看，点赞