Scholarlink

<p>In practical scenarios, network structures often encounter difficulties in acquiring complete network data, which are mainly caused by constraints such as statistical errors and the settings of access rights. Consequently, network completion technology holds great research significance in domains including graph mining and knowledge graph reasoning. In this paper, aiming at the situation where some links of the missing nodes in the network are lost, a missing node detection method named MND-C, which is based on self-supervised contrastive learning, is proposed. This method incorporates a node representation learning framework grounded in self-supervised contrastive learning. It learns node representations by maximizing the similarity of two positive nodes, eliminating the necessity of constructing a specific neural network architecture. Furthermore, this method samples the neighborhood subgraph of the central node as the input sample for the model, and leverages the local neighborhood information of nodes to carry out the missing node detection task. Experiments performed on several real-world datasets demonstrate that MND-C is both highly feasible and effective&#xa0;in the task of missing node detection.</p>

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

MND-C: a contrastive self-supervised framework for missing node detection in graphs

  • 期刊论文

  • 发表日期：

    2025年10月6日

查看全文
  • Xiangyi Kong,
  • Lixin Zhou,
  • Xuan Yao,
  • Weiyi Yue

摘要

In practical scenarios, network structures often encounter difficulties in acquiring complete network data, which are mainly caused by constraints such as statistical errors and the settings of access rights. Consequently, network completion technology holds great research significance in domains including graph mining and knowledge graph reasoning. In this paper, aiming at the situation where some links of the missing nodes in the network are lost, a missing node detection method named MND-C, which is based on self-supervised contrastive learning, is proposed. This method incorporates a node representation learning framework grounded in self-supervised contrastive learning. It learns node representations by maximizing the similarity of two positive nodes, eliminating the necessity of constructing a specific neural network architecture. Furthermore, this method samples the neighborhood subgraph of the central node as the input sample for the model, and leverages the local neighborhood information of nodes to carry out the missing node detection task. Experiments performed on several real-world datasets demonstrate that MND-C is both highly feasible and effective in the task of missing node detection.