LLaGA: Large Language and Graph Assistant

TL;DR

This work presents LLaGA, a framework that integrates Large Language Models with graph-structured data by translating graphs into node embedding sequences via structure-aware templates. A single trainable projector aligns these embeddings with the LLM's token space, enabling multi-task learning and zero-shot generalization across diverse graph datasets. The approach achieves competitive or superior performance on node classification, link prediction, and node description across four datasets, while also providing interpretable explanations for node embeddings. Ablation studies and zero-shot experiments demonstrate the benefits of the two templates and the framework's robustness to unseen data and tasks, signaling practical impact for broad graph analytics with LLMs. Overall, LLaGA offers a versatile, interpretable, and scalable route to leveraging LLMs for graph tasks without extensive task-specific fine-tuning.

Abstract

Graph Neural Networks (GNNs) have empowered the advance in graph-structured data analysis. Recently, the rise of Large Language Models (LLMs) like GPT-4 has heralded a new era in deep learning. However, their application to graph data poses distinct challenges due to the inherent difficulty of translating graph structures to language. To this end, we introduce the Large Language and Graph Assistant (LLaGA), an innovative model that effectively integrates LLM capabilities to handle the complexities of graph-structured data. LLaGA retains the general-purpose nature of LLMs while adapting graph data into a format compatible with LLM input. LLaGA achieves this by reorganizing graph nodes to structure-aware sequences and then mapping these into the token embedding space through a versatile projector. LLaGA excels in versatility, generalizability and interpretability, allowing it to perform consistently well across different datasets and tasks, extend its ability to unseen datasets or tasks, and provide explanations for graphs. Our extensive experiments across popular graph benchmarks show that LLaGA delivers outstanding performance across four datasets and three tasks using one single model, surpassing state-of-the-art graph models in both supervised and zero-shot scenarios. Our code is available at \url{https://github.com/VITA-Group/LLaGA}.

Figures (1)

• Figure 1: Illustration of LLaGA framework and its prompt design paradigm.