Ant Group Introduces Large Graph Model (LGM) Integrating Graph Computing and Large Language Models

Bill Gates described generative AI as the most important technological revolution in the past 40 years, highlighting the industry's focus on advancing scientific research and innovative applications as intelligent systems emerge.

On September 7, at the 2023 Bund Conference sub‑forum “Next‑Generation Data Infrastructure – Exploring Graph Intelligence Applications and Development,” Ant Group presented the “Large Graph Model” (LGM), which combines graph computing, graph learning, and large language models (LLMs). By leveraging the generative capabilities of LLMs and the relational analysis strengths of graph computing, LGM offers more intuitive, comprehensive information presentation and precise insights for solving massive, complex digital‑application challenges. Ant has completed the first phase, termed “generative heterogeneous graph augmentation,” and the related research paper was accepted at the WWW 2023 conference.

Graph computing is a powerful data‑processing technique that addresses relational difficulties in complex networks, with applications ranging from financial fraud detection and weather forecasting to drug discovery and brain‑inspired research, earning it the reputation of being the “bull nose” of artificial intelligence. Large models are viewed as the most promising path toward artificial general intelligence, achieving performance comparable to or surpassing humans in certain tasks.

Why combine cutting‑edge technologies? While LLMs can infer hidden relationships, they cannot directly render relationship graphs; representing data as graphs clarifies the chains of connections, making analysis more understandable. Integrating LLM reasoning with graph computation is akin to attaching a supercomputer to the human brain, enhancing overall capability.

Ant's research focuses on two main contributions. First, it uses LLMs to enrich graph data through a process called “generative heterogeneous graph augmentation,” where LLMs generate new data points beyond ordinary context‑dependent models. Second, it employs prompts to guide models in learning and discovering specific data characteristics—for example, prompting the model to generate samples of groups that have defaulted more than three times within a year—thereby accelerating data analysis and feature discovery.

Ant Group is a leader in graph computing, co‑developing the TuGraph platform with Tsinghua University, breaking LDBC SNB world‑record benchmarks three times, winning the “Leading Technology Achievement” award at the World Internet Conference in 2021, and being named an IDC MarketScape China Graph Database Market Leader. The “Large Graph Model” initiative began in early 2022 and gained feasibility validation with the surge of large‑model breakthroughs at the end of 2022.

The forum also featured insights from international experts: Professor M. Tamer Özsu (University of Waterloo) discussed streaming graph computing; Professor Chen Huajun (Zhejiang University) addressed opportunities and challenges of knowledge processing in the era of large models; Deputy Director Chen Hongyang (Zhejiang Lab) presented the latest research on graph scientific computing for biomedicine; and Li Yazhou (Co‑founder of Machine Heart) examined speculative issues and practical considerations of current large models.

Overall, the Bund Graph Intelligence Forum gathered diverse perspectives on the convergence of artificial intelligence and graph computing, outlining a crucial development trajectory for graph‑intelligent technologies.