Scholarlink

<p>Semi-supervised clustering (SSC) has received increasing attention in various real-world tasks. Typically, SSC relies on labeled data to supervise the clustering process. However, noisy or insufficient labeled data can mislead clustering or limit the ability to identify optimal solutions. To address these issues, this paper presents an adjusted compact-cluster assumption and thereby proposes a density-based semi-supervised hierarchical clustering (DBSSHC) method which aims to group data into a number of compact clusters. Both labeled data and density variations within clusters are employed to assess the compactness of clusters and guide the top-down clustering process. Especially, a density-based local clustering algorithm is proposed to identify low-density separations and further split uncompact clusters. In summary, this paper proposes a novel SSC algorithm by combining hierarchical clustering and density-based clustering. It not only compensates the deficiency of labeled data but also decreases the impact of noise among labeled data. The proposed method is validated on a variety of benchmark datasets. Measured by NMI, Rn, Accuracy, and Purity, DBSSHC achieves significant superiority over other state-of-the-art SSC methods and yields the best results on more datasets. These experimental results demonstrate the effectiveness of the proposed method. The source code of DBSSHC is available at <a href="https://codeocean.com/capsule/9190794/tree">https://codeocean.com/capsule/9190794/tree</a>.</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

Semi-supervised hierarchical clustering via low-density separation

  • 期刊论文

  • 发表日期：

    2026年3月9日

查看全文
  • Zhen Jiang,
  • Haolei Qin,
  • Zeyu Feng,
  • Haizhen Guan

摘要

Semi-supervised clustering (SSC) has received increasing attention in various real-world tasks. Typically, SSC relies on labeled data to supervise the clustering process. However, noisy or insufficient labeled data can mislead clustering or limit the ability to identify optimal solutions. To address these issues, this paper presents an adjusted compact-cluster assumption and thereby proposes a density-based semi-supervised hierarchical clustering (DBSSHC) method which aims to group data into a number of compact clusters. Both labeled data and density variations within clusters are employed to assess the compactness of clusters and guide the top-down clustering process. Especially, a density-based local clustering algorithm is proposed to identify low-density separations and further split uncompact clusters. In summary, this paper proposes a novel SSC algorithm by combining hierarchical clustering and density-based clustering. It not only compensates the deficiency of labeled data but also decreases the impact of noise among labeled data. The proposed method is validated on a variety of benchmark datasets. Measured by NMI, Rn, Accuracy, and Purity, DBSSHC achieves significant superiority over other state-of-the-art SSC methods and yields the best results on more datasets. These experimental results demonstrate the effectiveness of the proposed method. The source code of DBSSHC is available at https://codeocean.com/capsule/9190794/tree.