Reproduce “Graph2text or Graph2token: A Perspective of Large Language Models for Graph Learning” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Shuo Yu, Yingbo Wang, Ruolin Li, Guchun Liu, Yanming Shen, Shaoxiong Ji, Bowen Li, Fengling Han and 2 more

This survey paper introduces a novel taxonomy for integrating Large Language Models (LLMs) with graph learning, categorizing methods into two paradigms: Graph2text and Graph2token. It identifies four core challenges—alignment, position, multi-level semantics, and context—that arise when transforming graph structures into formats LLMs can process, providing a structured framework for understanding this interdisciplinary field.

Key insights

  • The integration of LLMs with graph learning offers a compelling way to overcome the 'semantic gap' and inherent structural challenges faced by traditional graph models.
  • A foundational taxonomy distinguishing between Graph2text and Graph2token transformations is crucial for understanding how LLMs interpret and process graph structures.
  • Four specific challenges—alignment, position, multi-level semantics, and context—are fundamental to effectively translating graph properties into a format consumable by LLMs.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.