Introduction

Graph-O1 is a new graph reasoning framework that combines Artificial Intelligence with Natural Language Knowledge, enabling models to reason over attributed graphs in a more accurate and interpretable way.

Key Features

• Integrates Monte Carlo Tree Search (MCTS) with end-to-end reinforcement learning to selectively explore and retrieve the most informative subgraph components.

• Frames the reasoning procedure as a multi-turn interaction between the agent and the graph environment, using a unified reward mechanism.

• Extensive experiments across multiple Large Language Models (LLMs) demonstrate that Graph-O1 consistently surpasses state-of-the-art baselines, producing answers that are more accurate, reliable, and interpretable.