LLM+GRAPH

LLM+Graph:

Large Language Models (LLMs) have seen rapid development such as ChatGPT and LLaMA, and have received tremendous attention from both industry and academia. While they have shown remarkable success over text data, recent works have demonstrated their limitations in reasoning with structured data including graphs. Graph-structured data is ubiquitous in the real world ranging from social and biological networks to financial transactions, knowledge bases, and transportation systems -- they permeate our daily lives. Therefore, understanding how to utilize graph data optimally with LLMs is a crucial research question. Recently, exploring the synergy between graphs and LLMs are attracting increasing interest in the data management and AI communities. On one hand, LLMs can be enhanced with graph computing techniques to provide answers with more contextualized facts, e.g., graph-based retrieval-augmented generation (graph RAG). On the other hand, downstream tasks, e.g., knowledge graph (KG) construction, graph data management and mining can also benefit by adopting LLMs, such as via LLM-graph DB (graph databases) and LLM-GNN (graph neural network) collaborations. It is, therefore, a timely opportunity to explore effective ways of interactions between LLMs and graph data. Our workshop “LLM+Graph” targets data management and data science researchers, aiming to inspect effective algorithms and systems to bringing LLMs, graph data management, and graph ML together in real applications.

We solicit unpublished papers discussing issues and successes under the broad category of Graph Data for LLMs, and LLMs for Graphs in the following areas (and beyond):

• Graph-enhanced Pre-training

• Graph-enhanced Fine-tuning

• Graph-enhanced Inference

• Graph-enhanced Refiners and Validators

• LLM-enhanced Graph Querying

• LLM-enhanced Graph Mining

• LLM-enhanced Graph Learning

Additionally, the paper must have a clear data management focus, e.g., discussing data management solution(s) such as (but not limited to):

• Data and Input Modeling for LLM+Graph

• Data Cleaning, Integration, and Augmentation with LLM+Graph

• Vector Data Management for LLM+Graph

• Accuracy and Consistency of LLM+Graph

• Explainability and Provenance of LLM+Graph

• Security and Privacy for LLM+Graph

• Benchmarking and Ground Truth for LLM+Graph

Submission Details:

We solicit and select five types of papers

• Survey Papers: these papers survey the related work in specific sub-areas and lay out the agenda for future work.

• Regular Research Papers: these are research papers with different flavors including foundations, algorithms, systems, information system architectures, experimental benchmarking, and applications. Papers with new/ late-breaking results are also welcome, which report the newest preliminary results about the most promising problems in the field.

• Vision Papers: these papers are devoted to discussing problems that we face currently and anticipate for the future.

• Demonstration Papers: these are software demonstration proposals, accompanied by short papers. The paper must describe the demonstrated system, user interface, options for user interactions, the system setup, and state the novelty and significance. We encourage providing the online link of a demonstration video, which is accessible by the reviewers.

• Extended Abstract Papers: if your research article has been accepted elsewhere, you are eligible to submit your work in the form of an extended abstract under the short paper category, while citing your previously published article. We encourage you to rephrase when needed or possible to avoid substantial verbatim text overlapping with past accepted/ published materials.

For survey, vision, demonstration, and extended abstract papers, you require to add [Survey], [Vision], [Demo], [Extended Abstract], respectively, next to the paper title, in both CMT submission form and also in the submitted paper pdf. We welcome the papers that fall under short papers of at most 6 pages and long papers up to 12 pages, including bibliography. Submissions must use the template at: vldb-workshop-style-master.zip. All submissions must be submitted in PDF through: https://cmt3.research.microsoft.com/LLMG2025/. Submissions will be reviewed in a single-blind manner, and all author names and affiliations should be included. Papers that do not follow the guidelines or are not within the scope of relevant topics will be desk rejected. We also expect that publications from DB venues, e.g., SIGMOD/VLDB/ICDE/EDBT etc. are cited. Submissions will be reviewed by at least three members of the Program Committee. All accepted papers will be published in VLDB workshop proceedings. The workshop will be in-person and at least one author of each accepted paper is required to register.

Accepted Paper List:

LLM-assisted Construction of the United States Legislative Graph
Francesco Cambria, Andrea Colombo

Scalable Graph-based Retrieval-Augmented Generation via Locality-Sensitive Hashing
Fangyuan Zhang, Zhengjun Huang, Yingli Zhou, Qingtian Guo, Wensheng Luo, Xiaofang Zhou

LLM-Hype: A Targeted Evaluation Framework for Hypernym-Hyponym Identification in Large Language Models
Qiu Ji, Pengfei Zhu, Haolei Zhu, Yang Sheng, Guilin Qi, Lianlong Wu, Kang Xu, Yuan Meng

Graph-Enhanced Large Language Models for Spatial Search [Vision]
Nicole Schneider, Kent O'Sullivan, Hanan Samet

xpSHACL: Explainable SHACL Validation using Retrieval-Augmented Generation and Large Language Models
Gustavo Publio, Jose Emilio Labra Gayo

Automatic Prompt Optimization for Knowledge Graph Construction: Insights from an Empirical Study
Nandana Mihindukulasooriya, Niharika DSouza, Faisal Chowdhury, Horst Samulowitz

Towards the Next Generation of Agent Systems: From RAG to Agentic AI [Vision]
Yingli Zhou, Shu Wang