News Text Classification Based on MLCNN and BiGRU Hybrid Neural Network

摘要

In the era of knowledge explosion, text classification is becoming increasingly crucial. At the same time, with the proposed Blockchain, it is of great research significance to actively explore the combination of Blockchain and AI, especially to apply text classification technology to the security classification of Blockchain technology. In this paper, we propose a hybrid neural network model (MLCNN & BiGRU-ATT) based on Multilayer Convolutional Neural Networks (MLCNN) and Bidirectional Gated Recurrent Unit (BiGRU) with Attention Mechanism in the news text classification field. GRU (Gate Recurrent Unit), a variant of LSTM (Long-Short Term Memory), has the natural advantages in processing time series tasks, which can readily capture the characteristics of text context information. Due to its prominent advantages in local feature extraction, CNN is also applied to NLP area, in which the researchers have made substantial progress. The experiment results reveal that our model has achieved higher accuracy on THUCNews dataset and Sougou news corpus classification.