This paper proposes a novel model called Attention-Induced IMputation Network (AIMNet) for the incomplete multi-view partial multi-label classification (iMvPMLC) task. AIMNet is designed to simultaneously extract label semantic features and instance embedding features, effectively handling both incomplete views and labels. The model addresses the challenge of missing data by approximating missing instances in the embedding space using cross-view joint attention, rather than recovering missing views in kernel or original feature space. This approach dynamically weights multi-view features based on confidence derived from joint attention during late fusion. Additionally, AIMNet incorporates a multi-view multi-label classification framework based on label-semantic feature learning, utilizing a statistical weak label correlation matrix and graph attention network to guide the learning process of label-specific features. The model is compatible with missing multi-view and partial multi-label data and has been validated through extensive experiments on five datasets, demonstrating its effectiveness and advancement. The AIMNet framework consists of two branches: an instance feature extraction branch and a label semantic extraction branch, which are fused in the embedding space to obtain predictions on each view. A multi-view late-fusion method is employed to fuse predictions of multiple views according to confidence, and missing instances are filled based on available inter-instance attention. The model's performance is evaluated using metrics such as ranking loss, average precision, Hamming loss, and area under the adaptation curve, showing superior results compared to existing methods. The AIMNet also introduces a confidence-based dynamic multi-view late fusion mechanism to improve the reliability of multi-view fusion. The model's effectiveness is further confirmed through ablation studies and time cost comparisons, demonstrating its adaptability and efficiency. The paper concludes that AIMNet is a promising approach for iMvPMLC, offering a robust solution to the challenges of incomplete multi-view and partial multi-label data.This paper proposes a novel model called Attention-Induced IMputation Network (AIMNet) for the incomplete multi-view partial multi-label classification (iMvPMLC) task. AIMNet is designed to simultaneously extract label semantic features and instance embedding features, effectively handling both incomplete views and labels. The model addresses the challenge of missing data by approximating missing instances in the embedding space using cross-view joint attention, rather than recovering missing views in kernel or original feature space. This approach dynamically weights multi-view features based on confidence derived from joint attention during late fusion. Additionally, AIMNet incorporates a multi-view multi-label classification framework based on label-semantic feature learning, utilizing a statistical weak label correlation matrix and graph attention network to guide the learning process of label-specific features. The model is compatible with missing multi-view and partial multi-label data and has been validated through extensive experiments on five datasets, demonstrating its effectiveness and advancement. The AIMNet framework consists of two branches: an instance feature extraction branch and a label semantic extraction branch, which are fused in the embedding space to obtain predictions on each view. A multi-view late-fusion method is employed to fuse predictions of multiple views according to confidence, and missing instances are filled based on available inter-instance attention. The model's performance is evaluated using metrics such as ranking loss, average precision, Hamming loss, and area under the adaptation curve, showing superior results compared to existing methods. The AIMNet also introduces a confidence-based dynamic multi-view late fusion mechanism to improve the reliability of multi-view fusion. The model's effectiveness is further confirmed through ablation studies and time cost comparisons, demonstrating its adaptability and efficiency. The paper concludes that AIMNet is a promising approach for iMvPMLC, offering a robust solution to the challenges of incomplete multi-view and partial multi-label data.