# A Survey of Large Language Models on Generative Graph Analytics: Query, Learning, and Applications

**Authors:** Wenbo Shang, Xin Huang  
**arXiv:** [2404.14809](https://arxiv.org/abs/2404.14809)  
**Date:** 2024-04-23  
**Summary page:** https://www.summarizepaper.com/en/paper/2404.14809v1/

## Key points

- Graphs are essential for representing complex relationships in society and nature, including social networks, transportation systems, financial structures, and biomedical networks.
- Large language models (LLMs) have transformed natural language processing (NLP) and multi-modal tasks with their exceptional generalization capabilities, particularly in addressing challenges related to graph tasks.
- The survey focuses on LLM applications in graph data analysis, highlighting the effectiveness of LLM models in various graph analytics tasks such as generative graph analytics (LLM-GGA), graph query processing (LLM-GQP), graph inference and learning (LLM-GIL), and graph-LLM-based applications.
- LLM-GQP integrates graph analytics techniques with LLM prompts to enhance functionalities like graph understanding and knowledge graph-based augmented retrieval.
- LLM-GIL leverages LLM capabilities for learning and reasoning over graphs through mechanisms such as graph learning, reasoning formation based on graphs, and enhancing overall graph representation.
- The survey explores how integrating different prompts into LLM models can effectively address various downstream tasks related to graphs.
- It provides insights into evaluating LLM models, benchmark datasets/tasks used for assessment purposes, along with a detailed analysis of pros and cons associated with employing these models.
- This interdisciplinary research area at the intersection of LLMs and graph analytics presents numerous open problems and future directions for exploration.

## Summary

In the realm of data representation, graphs serve as a fundamental model to depict intricate relationships among various entities in society and nature. They encompass social networks, transportation systems, financial structures, and biomedical networks. The emergence of large language models (LLMs) has revolutionized the landscape of natural language processing (NLP) and multi-modal tasks by showcasing remarkable generalization capabilities. These advanced models excel in addressing the challenges associated with graph tasks, surpassing traditional graph learning models. This comprehensive survey delves into existing studies on LLM applications in graph data analysis. It sheds light on the diverse range of graph analytics tasks that can be effectively tackled using sophisticated LLM models. The investigation highlights key areas within LLM-based generative graph analytics (LLM-GGA), categorizing them into LLM-based graph query processing (LLM-GQP), LLM-based graph inference and learning (LLM-GIL), and graph-LLM-based applications. Within the domain of LLM-GQP, emphasis is placed on integrating graph analytics techniques with LLM prompts to enhance functionalities such as graph understanding and knowledge graph (KG) based augmented retrieval. On the other hand, LLM-GIL focuses on leveraging LLM capabilities for learning and reasoning over graphs through mechanisms like graph learning, reasoning formation based on graphs, and enhancing overall graph representation. The survey also explores how various prompts integrated into LLM models can effectively address different downstream tasks related to graphs. Additionally, it provides insights into evaluating LLM models, benchmark datasets/tasks used for assessment purposes, along with a detailed analysis of pros and cons associated with employing these models. Furthermore, this interdisciplinary research area at the intersection of LLMs and graph analytics presents numerous open problems and future directions that warrant exploration. By bridging the gap between advanced language modeling techniques and complex network analysis methodologies, this study paves the way for innovative advancements in understanding and harnessing the power of interconnected data structures within diverse domains.