Label-Based Convolutional Neural Network for Text Classification
Chen Wang, Changgeng Tan · 2021
The neural network models based on word embedding have achieved remarkable results in text classification. Even so, these models hardly consider that the importance of each word and labels for text classification is beneficial to obtain informative text representation. The attention mechanisms usually are used to measure the weights of words to improve predictive performance, but we attempt to achieve the same goal in a simple way. Since that word embedding can capture semantic regularities between words. we introduce a text representation based on label by embedding each label and the word vectors in the same space in this paper. In this label-based text representation, each word has weight information of the number of classes, which play an important role in the final performance. So we proposed a labelbased convolutional neural network (LBCNN) to obtain the importance of different word in the label-based text sequence and the most influential semantic features in the word vector respectively. The experimental results show that our proposed method outperforms the state-of-art methods on the several large text classification datasets.