An improved text classification method based on convolutional neural networks

Abstract

To improve the classification accuracy of complaint work order text data, a deep learning-based classification method is designed. The word vector of this paper uses word2vec. Although word2vec represents the semantic richness of the words, it ignores the semantic information of the local words of the sentence. The word vector using a combination of n-gram and word2vec is both semantically rich and takes into account the local word order. In terms of the classification model, a combination of attention and CNN is used to consider both global and local features. After several sets of comparative experiments, the proposed algorithm for text classification on a company's complaint text effectively improves the accuracy rate. The accuracy rate is better than other algorithms reaching more than 90%.