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 Encodings2. GLAD: Bidirectional Structure-Attribute Alignment via Latent Graph Diffusion Models3. OpenMAG: A Comprehensive Benchmark for Multimodal-Attributed Graph4. Toward Effective Multimodal Graph Foundation Model: A Divide-and-Conquer Based Approach5. Structured Multi-modal Graph Disentanglement for Psychiatric Diagnosis6. MDGMIX: Boundary-Aware Subgraph Mixing for Multi-Domain Graph Pre-Training7. Is Fixing Schema Graphs Necessary? Full-Resolution Graph Structure Learning for Relational Deep Learning8. What Makes a Desired Graph for Relational Deep Learning?9. CCLRec: Consensus-driven Contrastive Learning for LLM-enhanced Graph Recommendation10. When LLMs Encounter Open-world Graph Learning: A Fresh View on Unlabeled Data Uncertainty11. Conformal Path Reasoning: Trustworthy Knowledge Graph Question Answering via Path-Level Calibration12. Graph is a Substrate Across Data Modalities13. GP2F: Cross-Domain Graph Prompting with Adaptive Fusion of Pre-trained Graph Neural Networks14. DTKG: Dual-Track Knowledge Graph-Verified Reasoning Framework for Multi-Hop QA15. Clustering as Reasoning: Akkk-Means Interpretation of Chain-of-Thought Graph Learning16. Large Language Models as Topological Thinkers: A Benchmark on Graph Persistent Homology17. Enhancing LLMs for Graph Tasks via Graph-aware LoRA Generation18. GraphPFN: A Prior-Data Fitted Network for Graph Node-Level Tasks19. GFMate: Empowering Graph Foundation Models with Pre-training-agnostic Test-time Prompt Tuning20. Structure-Centric Graph Foundation Model via Geometric Bases21. A Graph Foundation Model with Cross-Modal Alignment and Modality-Aware Expert Fusion for Multi-Modal Graphs22. Learning Graph Foundation Models on Riemannian Graph-of-Graphs23. When Do Graph Foundation Models Transfer? A Data-Centric Theory24. Message Tuning Outshines Graph Prompt Tuning: A Prismatic Space Perspective25. Graph-GRPO: Training Graph Flow Models with Reinforcement Learning26. Position: Graph Condensation Needs a Reset—Move Beyond Full-dataset Training and Model-Dependence27. DiP-G: Discrete Prompting for Graph Neural Networks28. GRASP: Graph Reasoning via Agentic Solving and Probing of LLMs29. Are Common Substructures Transferable? Understanding Transferability in Graph Pretraining under Riemannian Geometry30. Bridging Structure and Semantics: Uncertainty-Modulated Dual-Path Diffusion for Robust Text-Attributed Graph Learning31. RSF-GLLM: Bridging the Semantic Gap in Multi-Hop Knowledge Graph QA via Recurrent Soft-Flow and Decoupled LLM Generation32. Backjump-on-Graph: Empowering LLMs with Reinforced Retrospective Exploration for Agentic KG Reasoning33. 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/63030arXivhttps://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/64650arXivhttp://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benchmark4 Toward Effective Multimodal Graph Foundation Model: A Divide-and-Conquer Based Approach链接https://icml.cc/virtual/2026/poster/64358arXivhttp://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5 Structured Multi-modal Graph Disentanglement for Psychiatric Diagnosis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/60613arXivhttps://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/61364arXivhttp://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/66111arXivhttp://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/63086arXivhttp://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GNN14 DTKG: Dual-Track Knowledge Graph-Verified Reasoning Framework for Multi-Hop QA链接https://icml.cc/virtual/2026/poster/66752arXivhttp://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: Akkk-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LoRa18 GraphPFN: A Prior-Data Fitted Network for Graph Node-Level Tasks链接https://icml.cc/virtual/2026/poster/66511arXivhttps://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/62244arXivhttp://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21 A Graph Foundation Model with Cross-Modal Alignment and Modality-Aware Expert Fusion for Multi-Modal Graphs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/63157arXivhttp://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/65744arXivhttp://arxiv.org/abs/2603.10395v1作者Baoheng Zhu ⋅ Deyu Bo ⋅ Delvin Zhang ⋅ Xiao Wang关键词图流模型GRPO26 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LLM29 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Agentic33 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