Label mining method and device fusing deep network and attention mechanism

The invention discloses a label mining method and device fusing a deep network and an attention mechanism, and the method comprises the steps: collecting original data, and extracting a high-order combination discrete feature of an original continuous feature through a tree model; coding the original discrete features according to the categories of the original discrete features to obtain coded features; inputting the high-order combination discrete feature and the coding feature into an embedding layer of a label model for processing to obtain an embedding vector; inputting the embedded vectors into an attention network of a label model to obtain the weight of each embedded vector, and updating the embedded vectors according to the weight of each embedded vector; and inputting the updated embedded vector into a label identification network of the label model in a cross manner to output a user label. According to the method, the user tags are mined by fusing the deep network and the attention mechanism, the a