<p>Multi-view clustering aims to enhance clustering performance by leveraging complementary information from different sources or modalities. However, in practical applications, the prevalent issue of missing views leads to information imbalance among samples and difficulties in cross-view correlation modeling, severely constraining clustering performance. While existing methods can effectively exploit consistency information across views, they often overlook the complementary information embedded within each individual view. Moreover, view completion and clustering optimization are typically performed separately, limiting overall performance. To address these challenges, this paper proposes a novel framework for multi-view clustering based on Cross-view Attention Enhancement and Entropy-constrained Contrastive Learning (CAEC-Net). This framework collaboratively encodes and enhances multi-view features through graph convolutional networks and a cross-view attention mechanism, achieving missing view completion and cross-view semantic alignment. Furthermore, we design an instance-level confidence-aware attention fusion mechanism to adaptively integrate multi-view information. During the clustering stage, entropy regularization constraints and cluster-level contrastive learning mechanisms are introduced to reduce clustering uncertainty and enhance inter-cluster separability. Additionally, a stage-wise optimization strategy is adopted to progressively guide the model from local contrastive alignment to global clustering consistency, ensuring training stability. Experimental results on multiple benchmark datasets demonstrate that CAEC-Net converges stably under missing view conditions and significantly outperforms existing state-of-the-art methods, validating its superior clustering performance, robustness, and generalization capability.</p>