Document Classification Based on semantic and Improved Convolutional Neural Network

Abstract

In order to improve the accuracy of text classification, we present a new convolution neural network model combining keyword and word-meaning transformation. We first preprocess the text and break words, and use sense labeling for semantic keywords and word-meaning transformation. and we divide the texts into two parts---word and word-meaning. Next, we use embedding layer to transform the word and word-meaning into corresponding word embedding. Then, we use improved convoluted neural network to train the model and extract higher-order features of text type data, and use multi-layer perceptron and SoftMax layer to classify the texts to predict the category of each text. Experimental results show that our document classification algorithm can get a high accuracy and the effect of classification of news topic detection gets well.