ICML 2026 | LLM×Graph论文总结[1]【图基础模型，文本属性图，多模态属性图，图对齐，图提示学习，关系深度学习

ICML 2026将在2026年7月6日---11日于韩国首尔（Seoul, South Korea）举行。本文总结了2026 ICML上有关LLM × Graph相关论文。如有疏漏，欢迎大家补充。

注：笔者将分为上下2篇推文来总结，本文主要涉及针对图任务本身的的论文。

本文Graph的Topic：图基础模型，文本属性图，多模态属性图，图对齐，图提示学习，关系深度学习，知识图谱问答等。

1. Graph Alignment for Benchmarking Graph Neural Networks and Learning Positional Encodings 2. GLAD: Bidirectional Structure-Attribute Alignment via Latent Graph Diffusion Models 3. OpenMAG: A Comprehensive Benchmark for Multimodal-Attributed Graph 4. Toward Effective Multimodal Graph Foundation Model: A Divide-and-Conquer Based Approach 5. Structured Multi-modal Graph Disentanglement for Psychiatric Diagnosis 6. MDGMIX: Boundary-Aware Subgraph Mixing for Multi-Domain Graph Pre-Training 7. Is Fixing Schema Graphs Necessary? Full-Resolution Graph Structure Learning for Relational Deep Learning 8. What Makes a Desired Graph for Relational Deep Learning? 9. CCLRec: Consensus-driven Contrastive Learning for LLM-enhanced Graph Recommendation 10. When LLMs Encounter Open-world Graph Learning: A Fresh View on Unlabeled Data Uncertainty 11. Conformal Path Reasoning: Trustworthy Knowledge Graph Question Answering via Path-Level Calibration 12. Graph is a Substrate Across Data Modalities 13. GP2F: Cross-Domain Graph Prompting with Adaptive Fusion of Pre-trained Graph Neural Networks 14. DTKG: Dual-Track Knowledge Graph-Verified Reasoning Framework for Multi-Hop QA 15. Clustering as Reasoning: A kkk-Means Interpretation of Chain-of-Thought Graph Learning 16. Large Language Models as Topological Thinkers: A Benchmark on Graph Persistent Homology 17. Enhancing LLMs for Graph Tasks via Graph-aware LoRA Generation 18. GraphPFN: A Prior-Data Fitted Network for Graph Node-Level Tasks 19. GFMate: Empowering Graph Foundation Models with Pre-training-agnostic Test-time Prompt Tuning 20. Structure-Centric Graph Foundation Model via Geometric Bases 21. A Graph Foundation Model with Cross-Modal Alignment and Modality-Aware Expert Fusion for Multi-Modal Graphs 22. Learning Graph Foundation Models on Riemannian Graph-of-Graphs 23. When Do Graph Foundation Models Transfer? A Data-Centric Theory 24. Message Tuning Outshines Graph Prompt Tuning: A Prismatic Space Perspective 25. Graph-GRPO: Training Graph Flow Models with Reinforcement Learning 26. Position: Graph Condensation Needs a Reset---Move Beyond Full-dataset Training and Model-Dependence 27. DiP-G: Discrete Prompting for Graph Neural Networks 28. GRASP: Graph Reasoning via Agentic Solving and Probing of LLMs 29. Are Common Substructures Transferable? Understanding Transferability in Graph Pretraining under Riemannian Geometry 30. Bridging Structure and Semantics: Uncertainty-Modulated Dual-Path Diffusion for Robust Text-Attributed Graph Learning 31. RSF-GLLM: Bridging the Semantic Gap in Multi-Hop Knowledge Graph QA via Recurrent Soft-Flow and Decoupled LLM Generation 32. Backjump-on-Graph: Empowering LLMs with Reinforced Retrospective Exploration for Agentic KG Reasoning 33. LLM-MatLogic: Executable Exchange Contracts for Knowledge-Graph Query Answering with Scoped Negation

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。

1 Graph Alignment for Benchmarking Graph Neural Networks and Learning Positional Encodings

链接：++https://icml.cc/virtual/2026/poster/63030++

arXiv ：++https://arxiv.org/abs/2505.13087++

作者：Adrien Lagesse ⋅ Marc Lelarge

关键词：benchmark，图对齐，位置编码

2 GLAD: Bidirectional Structure-Attribute Alignment via Latent Graph Diffusion Models

链接：++https://icml.cc/virtual/2026/poster/61411++

作者：Jiankai Zuo ⋅ Yu Zhang ⋅ Yang Zhang ⋅ Zihao Yao ⋅ YAYING ZHANG

关键词：对齐，潜在图扩散模型

3 OpenMAG: A Comprehensive Benchmark for Multimodal-Attributed Graph

链接：++https://icml.cc/virtual/2026/poster/64650++

arXiv ：++http://arxiv.org/abs/2602.05576v1++

代码：++https://github.com/YUKI-N810/OpenMAG++

作者：Chenxi Wan ⋅ Xunkai Li ⋅ Yilong Zuo ⋅ Haokun Deng ⋅ Sihan Li ⋅ Bowen Fan ⋅ Hongchao Qin ⋅ Rong-Hua Li ⋅ Guoren Wang

关键词：多模态属性图，benchmark

4 Toward Effective Multimodal Graph Foundation Model: A Divide-and-Conquer Based Approach

链接：++https://icml.cc/virtual/2026/poster/64358++

arXiv ：++http://arxiv.org/abs/2602.04116v1++

作者：Sicheng Liu ⋅ Xunkai Li ⋅ Daohan Su ⋅ Ru Zhang ⋅ Hongchao Qin ⋅ Rong-Hua Li ⋅ Guoren Wang

关键词：多模态图基础模型

链接：++https://icml.cc/virtual/2026/poster/62853++

作者：Hongyu Shi ⋅ Kaizhong Zheng ⋅ WS Zhai ⋅ Shuai Jiang ⋅ Liangjun Chen ⋅ Badong Chen

关键词：多模态图解耦

6 MDGMIX: Boundary-Aware Subgraph Mixing for Multi-Domain Graph Pre-Training

链接：++https://icml.cc/virtual/2026/poster/65998++

作者：Ziyu Zheng ⋅ Yaming Yang ⋅ Ziyu Guan ⋅ Wei Zhao ⋅ Xinyan Huang

关键词：多域图预训练，子图混合

7 Is Fixing Schema Graphs Necessary? Full-Resolution Graph Structure Learning for Relational Deep Learning

链接：++https://icml.cc/virtual/2026/poster/66492++

作者：Yi Huang ⋅ Qingyun Sun ⋅ Jia Li ⋅ Xingcheng Fu ⋅ Jianxin Li

关键词：关系深度学习（RDL），图结构学习

8 What Makes a Desired Graph for Relational Deep Learning?

链接：++https://icml.cc/virtual/2026/poster/65162++

作者：Yao Cheng ⋅ Siqiang Luo

关键词：关系深度学习（RDL），图结构学习

9 CCLRec: Consensus-driven Contrastive Learning for LLM-enhanced Graph Recommendation

链接：++https://icml.cc/virtual/2026/poster/65594++

作者：Ting Guo ⋅ Dongyu Pei ⋅ Litiao Qiu ⋅ Xiaoying Liao ⋅ KE LIANG ⋅ Peng Song ⋅ Pinle Qin

关键词：基于图的推荐，对比学习，LLM增强

10 When LLMs Encounter Open-world Graph Learning: A Fresh View on Unlabeled Data Uncertainty

链接：++https://icml.cc/virtual/2026/poster/60613++

arXiv ：++https://arxiv.org/abs/2505.13989++

作者：Yanzhe Wen ⋅ Xunkai Li ⋅ Qi Zhang ⋅ Lei Zhu ⋅ Guang Zeng ⋅ Zhihan Zhang ⋅ Rong-Hua Li ⋅ Guoren Wang

关键词：开放世界图学习，未标记数据不确定性

11 Conformal Path Reasoning: Trustworthy Knowledge Graph Question Answering via Path-Level Calibration

链接：++https://icml.cc/virtual/2026/poster/61364++

arXiv ：++http://arxiv.org/abs/2605.08077v1++

作者：Shuhang Lin ⋅ Chuhao Zhou ⋅ Xiao Lin ⋅ Zihan Dong ⋅ Kuan Lu ⋅ Zhencan Peng ⋅ Jie Yin ⋅ Dimitris Metaxas

关键词：可信知识图谱问答，路径校准，共形路径推理

12 Graph is a Substrate Across Data Modalities

链接：++https://icml.cc/virtual/2026/poster/66111++

arXiv ：++http://arxiv.org/abs/2601.22384v1++

作者：Ziming Li ⋅ Xiao-Ming Wu ⋅ Zehong Wang ⋅ Jiazheng Li ⋅ Yijun Tian ⋅ Jinhe Bi ⋅ Yunpu Ma ⋅ Yanfang Ye ⋅ Chuxu Zhang

关键词：跨模态迁移

13 GP2F: Cross-Domain Graph Prompting with Adaptive Fusion of Pre-trained Graph Neural Networks

链接：++https://icml.cc/virtual/2026/poster/63086++

arXiv ：++http://arxiv.org/abs/2602.11629v1++

作者：Dongxiao He ⋅ Wenxuan Sun ⋅ Yongqi Huang ⋅ Jitao Zhao ⋅ Di Jin

关键词：跨域图提示学习，预训练GNN

14 DTKG: Dual-Track Knowledge Graph-Verified Reasoning Framework for Multi-Hop QA

链接：++https://icml.cc/virtual/2026/poster/66752++

arXiv ：++http://arxiv.org/abs/2510.16302v1++

作者：Changhao Wang ⋅ Yanfang Liu ⋅ Xinxin Fan ⋅ Lanzhi Zhou ⋅ Ao Tian ⋅ Yunfeng Lu

关键词：双轨知识图谱，多跳问答

15 Clustering as Reasoning: A kkk-Means Interpretation of Chain-of-Thought Graph Learning

链接：++https://icml.cc/virtual/2026/poster/63141++

作者：Xuanting Xie ⋅ Zhaochen Guo ⋅ Bingheng Li ⋅ Xingtong Yu ⋅ Zhifei Liao ⋅ zhao kang ⋅ Yuan Fang

关键词：思维链，图表示学习

16 Large Language Models as Topological Thinkers: A Benchmark on Graph Persistent Homology

链接：++https://icml.cc/virtual/2026/poster/63640++

作者：Hao Li ⋅ Hao Wan ⋅ Yixue Huang ⋅ Yuzhou Chen ⋅ Yulia Gel ⋅ Hao Jiang

关键词：拓扑理论，‌持续同调

17 Enhancing LLMs for Graph Tasks via Graph-aware LoRA Generation

链接：++https://icml.cc/virtual/2026/poster/65661++

作者：Junshu Sun ⋅ Wanxing Chang ⋅ Qingming Huang ⋅ Shuhui Wang

关键词：图感知的LoRa

18 GraphPFN: A Prior-Data Fitted Network for Graph Node-Level Tasks

链接：++https://icml.cc/virtual/2026/poster/66511++

arXiv ：++https://arxiv.org/abs/2509.21489++

作者：Dmitry Eremeev ⋅ Oleg Platonov ⋅ Gleb Bazhenov ⋅ Artem Babenko ⋅ Liudmila Prokhorenkova

关键词：图基础模型

19 GFMate: Empowering Graph Foundation Models with Pre-training-agnostic Test-time Prompt Tuning

链接：++https://icml.cc/virtual/2026/poster/65117++

作者：Yan Jiang ⋅ Ruihong Qiu ⋅ Zi Huang

关键词：图基础模型，测试时提示调优

20 Structure-Centric Graph Foundation Model via Geometric Bases

链接：++https://icml.cc/virtual/2026/poster/62244++

arXiv ：++http://arxiv.org/abs/2605.08689v1++

代码：++https://github.com/Xd-He/SCGFM++

作者：Xiaodong He ⋅ Haolan He ⋅ Ruiyi Fang ⋅ Ming Sun ⋅ zhao kang

关键词：图基础模型，结构为中心，几何基

链接：++https://icml.cc/virtual/2026/poster/62088++

作者：Dongxiao He ⋅ AnKang Yang ⋅ Jitao Zhao ⋅ Di Jin

关键词：图基础模型，跨模态对齐，专家聚合

22 Learning Graph Foundation Models on Riemannian Graph-of-Graphs

链接：++https://icml.cc/virtual/2026/poster/63157++

arXiv ：++http://arxiv.org/abs/2605.09993v1++

代码：++https://github.com/USTC-DataDarknessLab/R-GFM++

作者：Haokun Liu ⋅ Zezhong Ding ⋅ Xike Xie

关键词：图基础模型，黎曼图中图

23 When Do Graph Foundation Models Transfer? A Data-Centric Theory

链接：++https://icml.cc/virtual/2026/poster/65422++

作者：Jiajun Zhu ⋅ Ying Chen ⋅ Peihao Wang ⋅ Yixuan He ⋅ Pan Li ⋅ Aditya Akella ⋅ Zhangyang "Atlas" Wang

关键词：图基础模型，数据中心

24 Message Tuning Outshines Graph Prompt Tuning: A Prismatic Space Perspective

链接：++https://icml.cc/virtual/2026/poster/65770++

作者：Yancheng Chen ⋅ Dun Ma ⋅ Shuai Zhang ⋅ Yang Liu ⋅ Xixun Lin ⋅ Xiangyu Zhao ⋅ Wenguo Yang ⋅ Wei Chen ⋅ Chuan Zhou

关键词：图基础模型，提示调优

25 Graph-GRPO: Training Graph Flow Models with Reinforcement Learning

链接：++https://icml.cc/virtual/2026/poster/65744++

arXiv ：++http://arxiv.org/abs/2603.10395v1++

作者：Baoheng Zhu ⋅ Deyu Bo ⋅ Delvin Zhang ⋅ Xiao Wang

关键词：图流模型，GRPO

26 Position: Graph Condensation Needs a Reset---Move Beyond Full-dataset Training and Model-Dependence

链接：++https://icml.cc/virtual/2026/poster/67213++

作者：Mridul Gupta ⋅ Samyak Jain ⋅ Vansh Ramani ⋅ HARIPRASAD KODAMANA ⋅ Sayan Ranu

关键词：图浓缩

27 DiP-G: Discrete Prompting for Graph Neural Networks

链接：++https://icml.cc/virtual/2026/poster/65482++

作者：Yumeng Zhao ⋅ Huiying Hu ⋅ Steve Wen ⋅ Junjie Shen ⋅ Bei Hua

关键词：图提示学习，小样本

28 GRASP: Graph Reasoning via Agentic Solving and Probing of LLMs

链接：++https://icml.cc/virtual/2026/poster/61718++

作者：Xiaojun Guo ⋅ Mingxue Tian ⋅ Chenheng Zhang ⋅ Xiaohan Wang ⋅ Jiajun Chai ⋅ Guojun Yin ⋅ Wei Lin ⋅ Yifei Wang ⋅ Yisen Wang

关键词：图推理，LLM

29 Are Common Substructures Transferable? Understanding Transferability in Graph Pretraining under Riemannian Geometry

链接：++https://icml.cc/virtual/2026/poster/66087++

作者：Li Sun ⋅ Zhenhao Huang ⋅ Yiding Wang ⋅ Qin Chen ⋅ Pietro Lió ⋅ Philip Yu

关键词：图预训练，迁移

30 Bridging Structure and Semantics: Uncertainty-Modulated Dual-Path Diffusion for Robust Text-Attributed Graph Learning

链接：++https://icml.cc/virtual/2026/poster/65665++

作者：Zhizhi Yu ⋅ Jiachen Liu ⋅ Qingyu Li ⋅ Dongxiao He ⋅ Di Jin

关键词：文本属性图，扩散模型，不确定性

31 RSF-GLLM: Bridging the Semantic Gap in Multi-Hop Knowledge Graph QA via Recurrent Soft-Flow and Decoupled LLM Generation

链接：++https://icml.cc/virtual/2026/poster/62235++

作者：Sambaran Bandyopadhyay ⋅ Ananth Muppidi

关键词：知识图谱多跳问答

32 Backjump-on-Graph: Empowering LLMs with Reinforced Retrospective Exploration for Agentic KG Reasoning

链接：++https://icml.cc/virtual/2026/poster/61995++

作者：Yunqi Zhang ⋅ Shiqi Yan ⋅ Zhenzhao Yuan ⋅ Wenrui Liang ⋅ Yangming Liu ⋅ Zhixiao Qi ⋅ Tianyi Zhang ⋅ Shijie Zhang ⋅ Wei-Qiang Zhang ⋅ Yongfeng Huang ⋅ Haixin Duan ⋅ Shuai Chen ⋅ Yubo Chen

关键词：知识图谱问答，Agentic

33 LLM-MatLogic: Executable Exchange Contracts for Knowledge-Graph Query Answering with Scoped Negation

链接：++https://icml.cc/virtual/2026/poster/64362++

作者：Dezhuang Miao ⋅ Xiaoming Zhang ⋅ Bo Zhang ⋅ Yibin Du ⋅ Xiang Li ⋅ Ruilin Zeng ⋅ Yirui QI

关键词：知识图谱问答，LLM