Scholarlink

<p>Multi-view clustering (MVC) methods can improve the clustering accuracy by integrating complementary information from multiple views. In recent years, multi-view clustering based on contrastive learning has become an effective MVC method. However, there are two limitations: (1) using shallow view fusion strategies, which is unable to effectively fuse features for multiple views and (2) using hard clustering method, which cannot effectively solve the uncertainty between objects and clusters. Therefore, we propose a contrastive multi-view clustering based on multi-head attention mechanisms and three-way decision (called TAC-MVC). Firstly, we use auto-encoder to extract the low-level features of each view, based on which contrastive learning is used to maximize the consistency of information among views and learn the high-level features and semantic labels of each view and secondly, introduce the multi-attention mechanism to fuse the high-level features of each view and obtain a more accurate multi-view fusion feature representation. Finally, we perform clustering based on the three-way decision for multi-view fusion features and semantic labels and obtain three soft clustering results for the core, fringe, and the trivial region, to solve the uncertainty between objects and clusters. Experimental results on different datasets validate the effectiveness of the method.</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

Contrastive multi-view clustering based on multi-head attention mechanisms and three-way decision

  • 期刊论文

  • 发表日期：

    2025年4月16日

查看全文
  • Mingyue Lu,
  • Yi Xu,
  • Wenke Chu,
  • Jiaye Gu,
  • Muyang Gao

摘要

Multi-view clustering (MVC) methods can improve the clustering accuracy by integrating complementary information from multiple views. In recent years, multi-view clustering based on contrastive learning has become an effective MVC method. However, there are two limitations: (1) using shallow view fusion strategies, which is unable to effectively fuse features for multiple views and (2) using hard clustering method, which cannot effectively solve the uncertainty between objects and clusters. Therefore, we propose a contrastive multi-view clustering based on multi-head attention mechanisms and three-way decision (called TAC-MVC). Firstly, we use auto-encoder to extract the low-level features of each view, based on which contrastive learning is used to maximize the consistency of information among views and learn the high-level features and semantic labels of each view and secondly, introduce the multi-attention mechanism to fuse the high-level features of each view and obtain a more accurate multi-view fusion feature representation. Finally, we perform clustering based on the three-way decision for multi-view fusion features and semantic labels and obtain three soft clustering results for the core, fringe, and the trivial region, to solve the uncertainty between objects and clusters. Experimental results on different datasets validate the effectiveness of the method.