Multi-label text classification based on sequence model

Abstract

In the multi-label text classification problem, the category labels are frequently related in the semantic space. In order to enhance the classification performance, using the correlation between labels and using the Encoder in the seq2seq model and the Decoder model with the attention mechanism, a multi-label text classification method based on sequence generation is proposed. First, the Encoder encodes the word vector in the text to form a semantic coding vector. Then, the LSTM neural network in the Decoder stage is utilized to process the dependency of the category label sequence to consider the correlation between the category labels, and the attention mechanism is added to calculate the probability of attention distribution. Highlight the effect of key input on the output, and improve the missing semantic problem caused by the input too long, and finally output the predicted label category. The experimental results show that our model is better than the existing model after considering the label correlation.