Presenters

Abstract: The convergence of large language models (LLMs) and graph mining offers unprecedented opportunities for extracting insights from massive and complex graph-structured data. This tutorial, Big Graph Mining with LLM: Techniques, Architecture, and Applications, aims to equip researchers, data scientists, and practitioners with a comprehensive understanding of how to leverage LLMs for scalable and intelligent graph mining.

We will begin by introducing fundamental concepts of graph mining — including node classification, link prediction, and community detection — and highlight the challenges of scaling to massive graphs (e.g., billions of nodes and edges). We will then explore how LLMs, with their powerful semantic reasoning and pattern recognition capabilities, can augment traditional graph mining tasks.

Recent research has explored the potential of large language models (LLMs) in solving graph problems by leveraging their reasoning and code-generation capabilities. Benchmarks such as GPT4Graph [6] and NLGraph [1] investigate LLMs’ ability to handle diverse graph reasoning tasks such as graph connectivity, path finding, and maximum flow framed in natural language, demonstrating that while models like GPT-4 can perform basic tasks, they struggle with complexity—an issue partly mitigated by techniques such as Build-a-Graph and Algorithmic Prompting. Reinforcing this finding, two diagnostic studies—Exploring the Limitations of Graph Reasoning in LLMs [3] and Revisiting the Graph Reasoning Ability of LLMs [4] show that LLMs consistently underperform on tasks involving larger or more complex graphs, such as grid structures, weighted traversals, and multi-step translations, even with few-shot or chain-of-thought prompting. prompting. Complementing this, the GRAIL [5] framework introduces a novel method for computing Graph Edit Distance (GED) and node alignment by prompting LLMs to generate executable Python code, thereby offering interpretable and domain-generalizable solutions. Unlike traditional neural approaches requiring NP-hard supervision, GRAIL employs self-evolving prompt tuning to iteratively improve performance without ground-truth labels. Together, these works highlight the growing promise of LLMs in bridging natural language understanding and structured graph computation through prompt engineering and code synthesis.
We will end the tutorial with a hands-on session on these topics for all LLM and Graph Mining enthusiasts.

1. Can Language Models Solve Graph Problems in Natural Language?
2. Large Language Model Guided Graph Clustering
3. Exploring the Limitations of Graph Reasoning in Large Language Models
4. Revisiting the Graph Reasoning Ability of Large Language Models: Case Studies in Translation, Connectivity, and Shortest Path.
5. GRAIL: Graph Edit Distance and Node Alignment using LLM-Generated Code
6. GPT4Graph: Can Large Language Models Understand Graph-Structured Data? An Empirical Evaluation and Benchmarking

Duration: Half day

Target Audience: Researchers and Practitioners in NLP and Graph Learning, Graduate Students in AI/ML.