Let's Ask GNN: Empowering Large Language Model for Graph In-Context Learning

2024-10-09

Conference on Empirical Methods in Natural Language Processing

AI Summary

Addressing the critical challenge that large language models (LLMs) cannot natively process text-attributed graphs (TAGs), this paper proposes AskGNN: a framework that leverages a graph neural network (GNN)-driven, learnable retrieval mechanism to dynamically identify the most discriminative labeled nodes during in-context learning (ICL), thereby guiding LLMs to perform graph classification, node classification, and link prediction. AskGNN is the first method to jointly model structural graph priors and supervision signals, enabling zero-shot, parameter-free adaptation of LLMs to graph tasks without fine-tuning. Extensive experiments across three fundamental graph learning tasks and seven state-of-the-art LLMs demonstrate that AskGNN consistently outperforms diverse baselines. The results validate a novel, efficient, and general paradigm for endowing LLMs with graph awareness—requiring no parameter updates while preserving task-agnostic applicability.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Natural Language Processing: (Large) Language Models
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

Textual Attributed Graphs (TAGs) are crucial for modeling complex real-world systems, yet leveraging large language models (LLMs) for TAGs presents unique challenges due to the gap between sequential text processing and graph-structured data. We introduce AskGNN, a novel approach that bridges this gap by leveraging In-Context Learning (ICL) to integrate graph data and task-specific information into LLMs. AskGNN employs a Graph Neural Network (GNN)-powered structure-enhanced retriever to select labeled nodes across graphs, incorporating complex graph structures and their supervision signals. Our learning-to-retrieve algorithm optimizes the retriever to select example nodes that maximize LLM performance on graph. Experiments across three tasks and seven LLMs demonstrate AskGNN's superior effectiveness in graph task performance, opening new avenues for applying LLMs to graph-structured data without extensive fine-tuning.

Problem

• Language Models
• Text Attribute Graphs
• Non-Text Data Handling

Innovation

• AskGNN
• Graph Neural Networks
• Text Attribute Graphs