Research on multi-label classification problems based on neural networks and label correlation

In multi-label learning, each category label should be determined by its own specific features. However, as the number of features increases, it's become more challenging to capture dependencies between multiple labels, which is detrimental to the multi-label classification problem. Therefore, a novel neural network for specific feature extraction with a multi-label learning framework is proposed. First, the neural network performs low-dimensional mapping of the original data and learns a potential subspace for multi-label classification through a nonlinear mapping. In addition, the introduction of label correlation factors in the classification model improves the model's classification accuracy. Experimental results and analysis on multiple multi-label datasets of different sizes validate the effectiveness and robustness of our proposed method.