Graph-R1: Incentivizing the Zero-Shot Graph Learning Capability in LLMs via Explicit Reasoning

2025-08-24

AI Summary

Graph Neural Networks (GNNs) suffer from fixed label spaces, while Large Language Models (LLMs) lack structural inductive biases for graph reasoning. Method: We propose the first zero-shot graph learning framework grounded in explicit chain-of-thought (CoT) reasoning. It unifies node, link, and graph classification tasks as text-based inference problems; introduces task-specific “rethink templates” to guide large reasoning models in performing multi-step logical deduction over linearized graph structures; and constructs a training dataset annotated with fine-grained, interpretable reasoning trajectories. Contribution/Results: This work pioneers an interpretable, explicit CoT mechanism for graphs, integrating structure-aware graph linearization with template-driven reasoning guidance. Experiments demonstrate that our method significantly outperforms existing state-of-the-art approaches under zero-shot settings across multiple benchmarks, achieving high accuracy, strong generalization to unseen labels and domains, and fully explainable predictions.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Reasoning under Uncertainty: Graphical Models
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning

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: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Generalizing to unseen graph tasks without task-pecific supervision remains challenging. Graph Neural Networks (GNNs) are limited by fixed label spaces, while Large Language Models (LLMs) lack structural inductive biases. Recent advances in Large Reasoning Models (LRMs) provide a zero-shot alternative via explicit, long chain-of-thought reasoning. Inspired by this, we propose a GNN-free approach that reformulates graph tasks--node classification, link prediction, and graph classification--as textual reasoning problems solved by LRMs. We introduce the first datasets with detailed reasoning traces for these tasks and develop Graph-R1, a reinforcement learning framework that leverages task-specific rethink templates to guide reasoning over linearized graphs. Experiments demonstrate that Graph-R1 outperforms state-of-the-art baselines in zero-shot settings, producing interpretable and effective predictions. Our work highlights the promise of explicit reasoning for graph learning and provides new resources for future research.

Problem

• Generalizing to unseen graph tasks without supervision
• Overcoming GNNs' fixed label space limitations
• Addressing LLMs' lack of structural inductive biases

Innovation

• GNN-free approach reformulating graph tasks
• Reinforcement learning framework with rethink templates
• Explicit reasoning over linearized graphs