A text classification method based on the merge-LSTM-CNN model

Abstract

An MLCNN (Merge-LSTM-CNN) based text classification model combining a Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN) is proposed due to CNN's deficiency in obtaining context dependency in text and feature loss of the deep neural network in terms of text character extraction. First, the vector representation of input text is realized through word embedding, then full-text semantics are integrated by extracting the local features of the text through a three-layer CNN. Meanwhile, LSTM is used to store the features of historical information in the text to obtain its context-related semantics. Second, input vectors are integrated with the outputs of CNN in each layer respectively to allow the reuse of original features.