Abstract: The results of multi-label text classification are largely affected by label correlation. In order to deal with the label correlation problem in more detail, a multi-label text classification method fused with attention mechanism is proposed. Firstly, after preprocessing the text and labels, two different embedding methods are used to extract features for label input; Secondly, the attention mechanism is used to process the information, for the text and label information, the self-attention mechanism is used for feature processing, the label attention mechanism and the interactive attention mechanism are used for dependency processing, and then the representations in two different states are obtained. Finally, when the text label information is combined twice, the text label information is fully represented and better label classification results are obtained. The experimental results show that the data presented by this method generally improve the precision and normalized discounted cumulative gain compared with the baseline method. Thus, the method can effectively fuse text and label information, alleviate label correlation problem, and help improve the performance of multi-label text classification tasks.