.png){kind=icon}
.png){kind=icon}
Return
SSLMVC: Self-supervised label-driven multi-view collaborative clustering
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neucom.2026.134492.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Self-supervised multi-view clustering has made significant progress, yet two challenges remain: sensitivity of single-granularity pseudo-labels to view noise, and imbalance in cross-view representation coupling causing self-supervised signal distortion. To address these issues, we propose a novel framework named Self-Supervised Label-driven Multi-View Collaborative clustering framework (SSLMVC). Specifically, SSLMVC promotes inter-view coupling by aligning samples across feature, semantic, and label spaces, and achieves decoupling through deep autoencoders with dual-level diversity constraints on features and labels. This multi-level coupling–decoupling design effectively balances view-specific details with shared semantics, mitigating the degradation of self-supervised learning. Additionally, inspired by multi-label learning, we develop a dual pseudo-label generation module that derives complementary label information from both view-level features and the global clustering structure. Drawing on knowledge distillation, these pseudo-labels are aligned and fused to reduce the sensitivity to noisy or inconsistent views. Extensive experiments on 10 public datasets and comparisons with 10 state-of-the-art baselines demonstrate the superior performance and robustness of our proposed approach. The code is available at https://github.com/ZhaoLabNote/The-official-implementation-of-SSLMVC .