BiLSTM model with Attention mechanism for multi-label news text classification

Abstract

Multi-label text classification is very important in text research. A news will have multiple classification labels, and the number of labels is uncertain, so how to accurately learn labels from text is difficult. In this paper, GloVe is used for vectorization, and BiLSTM is used to learn the context, Use the Attention mechanism to focus on important words in the sentence, and finally focal loss function is used as the loss function. The experimental results show that the accuracy of this method reaches 84,43%% and AUR rate is 95.98%, which are higher than the evaluation scores of other five models.