Scholarlink

Heterogeneous graph neural networks (HGNNs) are emerging techniques with remarkable capabilities in handling heterogeneous graphs. Self-supervised HGNN models have garnered significant attention. Contrastive learning, due to its excellent performance, is widely employed in heterogeneous graph representation learning. However, existing methods may lead to the loss of original topology and feature information, and there are limitations in positive and negative sample selection. To address these issues, we present a self-supervised learning approach, HetGNN-TF, on heterogeneous graph neural networks via topology and feature reconstruction. We specifically reconstruct semantic structure using an encoder-decoder to guide the model to learn original structural information. We also construct a feature similarity graph to reconstruct node features through specific encoders, guiding them to fully grasp feature information. Additionally, we design an adaptive sample selection mechanism for iterative optimization of samples. Experimental results demonstrate the superior performance of our approach over existing state-of-the-art methods. The source code and data are available on GitHub ( https://github.com/LiuXMaa/HetGNN-TF.git ).

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

HetGNN-TF: Self-supervised Learning on Heterogeneous Graph Neural Network via Topology and Feature Reconstruction

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Chao Li,
  • Xinming Liu,
  • Jinhu Fu,
  • Zhongying Zhao,
  • Qingtian Zeng

摘要

Heterogeneous graph neural networks (HGNNs) are emerging techniques with remarkable capabilities in handling heterogeneous graphs. Self-supervised HGNN models have garnered significant attention. Contrastive learning, due to its excellent performance, is widely employed in heterogeneous graph representation learning. However, existing methods may lead to the loss of original topology and feature information, and there are limitations in positive and negative sample selection. To address these issues, we present a self-supervised learning approach, HetGNN-TF, on heterogeneous graph neural networks via topology and feature reconstruction. We specifically reconstruct semantic structure using an encoder-decoder to guide the model to learn original structural information. We also construct a feature similarity graph to reconstruct node features through specific encoders, guiding them to fully grasp feature information. Additionally, we design an adaptive sample selection mechanism for iterative optimization of samples. Experimental results demonstrate the superior performance of our approach over existing state-of-the-art methods. The source code and data are available on GitHub ( https://github.com/LiuXMaa/HetGNN-TF.git ).