Adnan KhanAdnanProduct & AI
San Francisco Bay Area
Get in touch

Source note3 min read

PromptGNN-sim: Deep Fusion and Alignment of GNN and LLMs for Text-Attributed Graph Learning

Source

arXiv

arxiv.org/abs/2606.30291v1

Read the source

Source

Plain-Language Brief

Researchers propose a new framework called PromptGNN-sim to improve the performance of Text-Attributed Graph Learning tasks. The method combines Graph Attention Network (GAT) and Language Model (LLM) in a bi-directional structure-semantic fusion framework for better interaction between modalities. During training, cross-modal contrastive learning and cross-attention are used to optimize both components.

What the Document Says

  • The paper introduces PromptGNN-sim, a bi-directional structure-semantic fusion framework for Text-Attributed Graph Learning tasks.
  • PromptGNN-sim uses a Graph Attention Network (GAT) to select semantically aware neighborhoods by combining structural attention and textual similarity.
  • The selected structural context is used to generate structure-aware prompts for an LLM, including the target node summary, label categories, and representative keywords from similar neighbors.
  • Bi-directional cross-modal contrastive learning and cross-attention are introduced during training to jointly optimize the GNN and LLM components.
  • The research evaluates accuracy, generalization, and robustness under various conditions such as cross-task transfer, cross-dataset generalization, and sparse perturbations.
  • Results show that PromptGNN-sim outperforms classical GNNs, LLMs, and recent GNN-LLM fusion methods.

Key Concepts Mentioned

  • Text-Attributed Graphs
  • Graph Attention Network (GAT)
  • Language Model (LLM)
  • PromptGNN-sim
  • structure-semantic fusion
  • cross-modal contrastive learning
  • cross-attention

Claims Made by the Document

  • Existing fusion methods limit deep interaction between modalities and weaken performance.
  • PromptGNN-sim addresses this issue by proposing a bi-directional structure-semantic fusion framework.
  • PromptGNN-sim outperforms classical GNNs, LLMs, and recent GNN-LLM fusion methods.

Evidence Used

  • The text describes the limitations of existing fusion methods and the introduction of PromptGNN-sim to address them.
  • Results from experiments on six public datasets are provided showing the outperformance of PromptGNN-sim.

Methods / Approach

  • Graph Attention Network (GAT)
  • Language Model (LLM)
  • bi-directional cross-modal contrastive learning
  • cross-attention

Limitations Stated by the Authors

  • The paper does not provide detailed information about the specific methods used for generating structure-aware prompts or optimizing the LLM component.

Limitations Noted by the Agent

Agent interpretation — not from the source.

  • The paper only evaluates performance on six public datasets. It is unknown how well PromptGNN-sim performs on other, unmentioned datasets.
  • It is also not stated in the source if PromptGNN-sim can handle complex graph structures that may be present in real-world applications.

Open Questions

  • How does PromptGNN-sim perform on other, unmentioned datasets?
  • Can PromptGNN-sim handle complex graph structures present in real-world applications?

Notes for Future Q&A

May include interpretation. Labeled where it does.

  • The authors propose a new framework for Text-Attributed Graph Learning tasks called PromptGNN-sim.
  • PromptGNN-sim combines a Graph Attention Network (GAT) and Language Model (LLM) in a bi-directional structure-semantic fusion framework.

Raw Abstract / Excerpt

Text-Attributed Graphs (TAGs) combine textual semantics with graph structure and are central to many graph learning tasks. However, existing fusion methods often treat text and structure as separate inputs in a shallow, one-way pipeline, which limits deep interaction between modalities and weakens performance under sparse connectivity or cross-graph generalisation. To address this issue, we propose PromptGNN-sim, a bi-directional structure-semantic fusion framework for collaborative GNN-LLM learning. PromptGNN-sim uses a Graph Attention Network (GAT) for semantically aware neighborhood selection by combining structural attention with textual similarity. The selected structural context is then used to generate structure-aware prompts for an LLM, including the target node summary, label categories, and representative keywords from similar neighbors. During training, bi-directional cross-modal contrastive learning and cross-attention are introduced to jointly optimize the GNN and LLM components. Experiments on six public datasets, including Cora, Pubmed, and WikiCS, evaluate accuracy, generalisation, and robustness under cross-task transfer, cross-dataset generalisation, and sparse perturbations. Results show that PromptGNN-sim outperforms classical GNNs, LLMs, and recent GNN-LLM fusion methods, demonstrating the effectiveness of interactive structure-semantic collaboration for text-attributed graph learning.