Scholarlink

<p>Contrastive learning (CL) is extensively applied in multi-view clustering by pulling positive samples closer and pushing negative samples apart. However, most existing deep contrastive multi-view clustering (DCMVC) methods are constrained to a single perspective, either at the feature or/and semantic level, ignoring the rich and crucial information between these two levels at intermediate granularities. Additionally, these methods concentrate solely on contrastive within the same level, overlooking potential interactions across different levels. This limitation restricts the representation capability of DCMVC, subsequently impacting clustering performance. To this end, we propose a novel framework dubbed CLMGC: cross-level interaction and multi-granularity contrastive learning for multi-view clustering. Specifically, we input the latent representations of each encoder into feature-level and semantic-level CL. In semantic-level CL, we propose a novel subdivision of the semantic level into two branches: fine-grained and coarse-grained semantic clusters. To further enhance the hierarchical richness of information, we introduce fine-grained dual contrastive mechanisms, including cross-level and self-instance contrast mechanisms, which connect feature-level with fine-grained semantic CL. This design enhances information transfer between different levels and significantly improves the discriminative capability of the fine-grained semantic cluster, thus optimizing the overall performance of CLMGC. Experimental results from six multi-view datasets demonstrate the superiority of the CLMGC algorithm compared with other state-of-the-art methods. The demo code of this work is publicly available at <a href="https://shanghui-deng.github.io">https://shanghui-deng.github.io</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

Cross-level interaction and multi-granularity contrastive learning for multi-view clustering

  • 期刊论文

  • 发表日期：

    2026年6月16日

查看全文
  • Shanghui Deng,
  • Chang Tang,
  • Xiao Zheng,
  • Yuanyuan Liu,
  • Kun Sun,
  • Xinwang Liu

摘要

Contrastive learning (CL) is extensively applied in multi-view clustering by pulling positive samples closer and pushing negative samples apart. However, most existing deep contrastive multi-view clustering (DCMVC) methods are constrained to a single perspective, either at the feature or/and semantic level, ignoring the rich and crucial information between these two levels at intermediate granularities. Additionally, these methods concentrate solely on contrastive within the same level, overlooking potential interactions across different levels. This limitation restricts the representation capability of DCMVC, subsequently impacting clustering performance. To this end, we propose a novel framework dubbed CLMGC: cross-level interaction and multi-granularity contrastive learning for multi-view clustering. Specifically, we input the latent representations of each encoder into feature-level and semantic-level CL. In semantic-level CL, we propose a novel subdivision of the semantic level into two branches: fine-grained and coarse-grained semantic clusters. To further enhance the hierarchical richness of information, we introduce fine-grained dual contrastive mechanisms, including cross-level and self-instance contrast mechanisms, which connect feature-level with fine-grained semantic CL. This design enhances information transfer between different levels and significantly improves the discriminative capability of the fine-grained semantic cluster, thus optimizing the overall performance of CLMGC. Experimental results from six multi-view datasets demonstrate the superiority of the CLMGC algorithm compared with other state-of-the-art methods. The demo code of this work is publicly available at https://shanghui-deng.github.io.