<p>Multi-view multi-label (MVML) learning present unique challenges due to the inherent complexity and dependencies among views and labels. To achieve MVML learning, most of the previous studies focus on using the shared subspace between different views to represent multi-view consensus information. However, these methods often assume all views contribute equally to the shared representation which leads to over-reliance on less informative views and under-utilization of more discriminative views. At the same time, the existing methods do not clearly capture the collaborative relationships between labels. In this paper, to address these challenges, we propose a Self-supervised Multi-view learning method with Attention mechanisms for Multi-label classification (SMAM). Firstly, we propose a view-masking self-supervised learning strategy, which enables efficient and robust learning of multi-view representations. Subsequently, cross-attention aligns label embeddings with multi-view representations to enhance semantic coherence. Finally, graph attention network is introduced to capture label dependencies. By integrating these components, our method achieves robust representation learning for both views and labels, facilitating accurate multi-label classification. Experimental results on five benchmark datasets demonstrate the effectiveness in MVML classification tasks.</p>