AgentGL: Towards Agentic Graph Learning with LLMs via Reinforcement Learning

2026-04-07

AI Summary

Existing agent frameworks struggle to effectively model the graph-topological dependencies inherent in real-world data, often treating external information as unstructured text. This work proposes Agentic Graph Learning (AGL), a novel paradigm that reframes graph learning as a synergistic process between topology-aware navigation and large language model (LLM) reasoning. We introduce AgentGL, the first reinforcement learning–based framework for AGL, which incorporates graph-native tools to enable multi-scale exploration, employs search constraints to regulate tool invocation, and features a graph-conditioned curriculum reinforcement learning strategy to achieve stable long-horizon training without step-by-step supervision. Evaluated on multiple textual attributed graph benchmarks, AgentGL substantially outperforms GraphLLM and GraphRAG, achieving performance gains of up to 17.5% in node classification and 28.4% in link prediction.

Abstract

Large Language Models (LLMs) increasingly rely on agentic capabilities-iterative retrieval, tool use, and decision-making-to overcome the limits of static, parametric knowledge. Yet existing agentic frameworks treat external information as unstructured text and fail to leverage the topological dependencies inherent in real-world data. To bridge this gap, we introduce Agentic Graph Learning (AGL), a paradigm that reframes graph learning as an interleaved process of topology-aware navigation and LLM-based inference. Specifically, we propose AgentGL, the first reinforcement learning (RL)-driven framework for AGL. AgentGL equips an LLM agent with graph-native tools for multi-scale exploration, regulates tool usage via search-constrained thinking to balance accuracy and efficiency, and employs a graph-conditioned curriculum RL strategy to stabilize long-horizon policy learning without step-wise supervision. Across diverse Text-Attributed Graph (TAG) benchmarks and multiple LLM backbones, AgentGL substantially outperforms strong GraphLLMs and GraphRAG baselines, achieving absolute improvements of up to 17.5% in node classification and 28.4% in link prediction. These results demonstrate that AGL is a promising frontier for enabling LLMs to autonomously navigate and reason over complex relational environments. The code is publicly available at https://github.com/sunyuanfu/AgentGL.

Problem

• Agentic Graph Learning
• Large Language Models
• Graph Reasoning
• Topological Dependencies
• Relational Environments

Innovation

• Agentic Graph Learning
• Reinforcement Learning
• Large Language Models
• Graph Reasoning
• Tool-Augmented LLMs