Less is More: Learning Graph Tasks with Just LLMs

2025-08-13

AI Summary

This work investigates whether large language models (LLMs) can learn and generalize graph-structured tasks without relying on graph neural networks (GNNs) or specialized graph encoders. It addresses three core questions: (i) whether LLMs can acquire fundamental graph reasoning capabilities, (ii) whether they generalize to unseen graph topologies and novel tasks, and (iii) how different methodological approaches compare. Method: We propose an instruction-guided chain-of-thought (CoT) training framework that combines textual serialization of graph structures with end-to-end instruction fine-tuning. Contribution/Results: Experiments demonstrate that small-scale LLMs achieve performance on par with—or surpassing—that of GNN-augmented hybrid models across diverse graph tasks, while exhibiting strong generalization to out-of-distribution graph topologies and task types. This study provides the first systematic empirical validation of pure language models for graph reasoning, establishing their feasibility and potential. It opens a new pathway toward lightweight, unified graph reasoning paradigms grounded solely in language modeling.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Reasoning under Uncertainty: Graphical Models

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
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

For large language models (LLMs), reasoning over graphs could help solve many problems. Prior work has tried to improve LLM graph reasoning by examining how best to serialize graphs as text and by combining GNNs and LLMs. However, the merits of such approaches remain unclear, so we empirically answer the following research questions: (1) Can LLMs learn to solve fundamental graph tasks without specialized graph encoding models?, (2) Can LLMs generalize learned solutions to unseen graph structures or tasks?, and (3) What are the merits of competing approaches to learn graph tasks? We show that even small LLMs can learn to solve graph tasks by training them with instructive chain-of-thought solutions, and this training generalizes, without specialized graph encoders, to new tasks and graph structures.

Problem

• Can LLMs solve graph tasks without specialized encoding models
• Do LLMs generalize learned solutions to unseen graph structures
• What are the merits of competing graph learning approaches

Innovation

• LLMs solve graph tasks without specialized encoders
• Training uses instructive chain-of-thought solutions
• Generalizes to new graph structures and tasks